Skip to content

P
PaddleOCR

weixin_41840029 / PaddleOCR
与 Fork 源项目一致

Fork自 PaddlePaddle / PaddleOCR

通知 1
Star 1
Fork 0
P
PaddleOCR
提交 253b8453 编写于 作者: qq_25193841's avatar qq_25193841
浏览文件
操作

Merge remote-tracking branch 'origin/dygraph' into dygraph

上级 7cad4817 bc999986
展开全部 显示空白变更内容
内联 并排
Showing with 9361 addition and 663 deletion
MANIFEST.in
浏览文件 @ 253b8453
include LICENSE.txt
include LICENSE
include README.md
recursive-include ppocr/utils *.txt utility.py logging.py
recursive-include ppocr/utils *.txt utility.py logging.py network.py
recursive-include ppocr/data/ *.py
recursive-include ppocr/postprocess *.py
recursive-include tools/infer *.py
......
configs/rec/ch_ppocr_v2.1/rec_chinese_lite_train_distillation_v2.1.yml 0 → 100644
浏览文件 @ 253b8453
Global:
debug: false
use_gpu: true
epoch_num: 800
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/rec_chinese_lite_distillation_v2.1
save_epoch_step: 3
eval_batch_step: [0, 2000]
cal_metric_during_train: true
pretrained_model:
checkpoints:
save_inference_dir:
use_visualdl: false
infer_img: doc/imgs_words/ch/word_1.jpg
character_dict_path: ppocr/utils/ppocr_keys_v1.txt
character_type: ch
max_text_length: 25
infer_mode: false
use_space_char: false
distributed: true
save_res_path: ./output/rec/predicts_chinese_lite_distillation_v2.1.txt
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Cosine
learning_rate: 0.0005
warmup_epoch: 5
regularizer:
name: L2
factor: 1.0e-05
Architecture:
name: DistillationModel
algorithm: Distillation
Models:
Student:
pretrained:
freeze_params: false
return_all_feats: true
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: small
small_stride: [1, 2, 2, 2]
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 64
Head:
name: CTCHead
mid_channels: 96
fc_decay: 0.00001
Teacher:
pretrained:
freeze_params: false
return_all_feats: true
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: small
small_stride: [1, 2, 2, 2]
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 64
Head:
name: CTCHead
mid_channels: 96
fc_decay: 0.00001
Loss:
name: CombinedLoss
loss_config_list:
- DistillationCTCLoss:
weight: 1.0
model_name_list: ["Student", "Teacher"]
key: head_out
- DistillationDMLLoss:
weight: 1.0
act: "softmax"
model_name_pairs:
- ["Student", "Teacher"]
key: head_out
- DistillationDistanceLoss:
weight: 1.0
mode: "l2"
model_name_pairs:
- ["Student", "Teacher"]
key: backbone_out
PostProcess:
name: DistillationCTCLabelDecode
model_name: ["Student", "Teacher"]
key: head_out
Metric:
name: DistillationMetric
base_metric_name: RecMetric
main_indicator: acc
key: "Student"
Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/
label_file_list:
- ./train_data/train_list.txt
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- RecAug:
- CTCLabelEncode:
- RecResizeImg:
image_shape: [3, 32, 320]
- KeepKeys:
keep_keys:
- image
- label
- length
loader:
shuffle: true
batch_size_per_card: 128
drop_last: true
num_sections: 1
num_workers: 8
Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data
label_file_list:
- ./train_data/val_list.txt
transforms:
- DecodeImage:
img_mode: BGR
channel_first: false
- CTCLabelEncode:
- RecResizeImg:
image_shape: [3, 32, 320]
- KeepKeys:
keep_keys:
- image
- label
- length
loader:
shuffle: false
drop_last: false
batch_size_per_card: 128
num_workers: 8
configs/table/table_mv3.yml 0 → 100755
浏览文件 @ 253b8453
Global:
use_gpu: true
epoch_num: 50
log_smooth_window: 20
print_batch_step: 5
save_model_dir: ./output/table_mv3/
save_epoch_step: 5
# evaluation is run every 400 iterations after the 0th iteration
eval_batch_step: [0, 400]
cal_metric_during_train: True
pretrained_model:
checkpoints:
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_words/ch/word_1.jpg
# for data or label process
character_dict_path: ppocr/utils/dict/table_structure_dict.txt
character_type: en
max_text_length: 100
max_elem_length: 500
max_cell_num: 500
infer_mode: False
process_total_num: 0
process_cut_num: 0
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
clip_norm: 5.0
lr:
learning_rate: 0.001
regularizer:
name: 'L2'
factor: 0.00000
Architecture:
model_type: table
algorithm: TableAttn
Backbone:
name: MobileNetV3
scale: 1.0
model_name: small
disable_se: True
Head:
name: TableAttentionHead
hidden_size: 256
l2_decay: 0.00001
loc_type: 2
Loss:
name: TableAttentionLoss
structure_weight: 100.0
loc_weight: 10000.0
PostProcess:
name: TableLabelDecode
Metric:
name: TableMetric
main_indicator: acc
Train:
dataset:
name: PubTabDataSet
data_dir: train_data/table/pubtabnet/train/
label_file_path: train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- ResizeTableImage:
max_len: 488
- TableLabelEncode:
- NormalizeImage:
scale: 1./255.
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: 'hwc'
- PaddingTableImage:
- ToCHWImage:
- KeepKeys:
keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
loader:
shuffle: True
batch_size_per_card: 32
drop_last: True
num_workers: 1
Eval:
dataset:
name: PubTabDataSet
data_dir: train_data/table/pubtabnet/val/
label_file_path: train_data/table/pubtabnet/PubTabNet_2.0.0_val.jsonl
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- ResizeTableImage:
max_len: 488
- TableLabelEncode:
- NormalizeImage:
scale: 1./255.
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: 'hwc'
- PaddingTableImage:
- ToCHWImage:
- KeepKeys:
keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
loader:
shuffle: False
drop_last: False
batch_size_per_card: 16
num_workers: 1
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MainActivity.java
浏览文件 @ 253b8453
......@@ -465,9 +465,13 @@ public class MainActivity extends AppCompatActivity {
}
public void btn_load_model_click(View view) {
if (predictor.isLoaded()){
tvStatus.setText("STATUS: model has been loaded");
}else{
tvStatus.setText("STATUS: load model ......");
loadModel();
}
}
public void btn_run_model_click(View view) {
Bitmap image =((BitmapDrawable)ivInputImage.getDrawable()).getBitmap();
......
deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java
浏览文件 @ 253b8453
......@@ -194,26 +194,25 @@ public class Predictor {
"supported!");
return false;
}
int[] channelStride = new int[]{width * height, width * height * 2};
int p = scaleImage.getPixel(scaleImage.getWidth() - 1, scaleImage.getHeight() - 1);
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
int color = scaleImage.getPixel(x, y);
int[] pixels=new int[width*height];
scaleImage.getPixels(pixels,0,scaleImage.getWidth(),0,0,scaleImage.getWidth(),scaleImage.getHeight());
for (int i = 0; i < pixels.length; i++) {
int color = pixels[i];
float[] rgb = new float[]{(float) red(color) / 255.0f, (float) green(color) / 255.0f,
(float) blue(color) / 255.0f};
inputData[y * width + x] = (rgb[channelIdx[0]] - inputMean[0]) / inputStd[0];
inputData[y * width + x + channelStride[0]] = (rgb[channelIdx[1]] - inputMean[1]) / inputStd[1];
inputData[y * width + x + channelStride[1]] = (rgb[channelIdx[2]] - inputMean[2]) / inputStd[2];
}
inputData[i] = (rgb[channelIdx[0]] - inputMean[0]) / inputStd[0];
inputData[i + channelStride[0]] = (rgb[channelIdx[1]] - inputMean[1]) / inputStd[1];
inputData[i+ channelStride[1]] = (rgb[channelIdx[2]] - inputMean[2]) / inputStd[2];
}
} else if (channels == 1) {
for (int y = 0; y < height; y++) {
for (int x = 0; x < width; x++) {
int color = inputImage.getPixel(x, y);
int[] pixels=new int[width*height];
scaleImage.getPixels(pixels,0,scaleImage.getWidth(),0,0,scaleImage.getWidth(),scaleImage.getHeight());
for (int i = 0; i < pixels.length; i++) {
int color = pixels[i];
float gray = (float) (red(color) + green(color) + blue(color)) / 3.0f / 255.0f;
inputData[y * width + x] = (gray - inputMean[0]) / inputStd[0];
}
inputData[i] = (gray - inputMean[0]) / inputStd[0];
}
} else {
Log.i(TAG, "Unsupported channel size " + Integer.toString(channels) + ", only channel 1 and 3 is " +
......
deploy/cpp_infer/include/utility.h
浏览文件 @ 253b8453
......@@ -44,6 +44,9 @@ public:
inline static size_t argmax(ForwardIterator first, ForwardIterator last) {
return std::distance(first, std::max_element(first, last));
}
static void GetAllFiles(const char *dir_name,
std::vector<std::string> &all_inputs);
};
} // namespace PaddleOCR
\ No newline at end of file
deploy/cpp_infer/readme.md
浏览文件 @ 253b8453
......@@ -77,7 +77,7 @@ opencv3/
#### 1.2.1 直接下载安装
* [Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/advanced_guide/inference_deployment/inference/build_and_install_lib_cn.html)上提供了不同cuda版本的Linux预测库，可以在官网查看并选择合适的预测库版本（*建议选择paddle版本>=2.0.1版本的预测库* ）。
* [Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html) 上提供了不同cuda版本的Linux预测库，可以在官网查看并选择合适的预测库版本（*建议选择paddle版本>=2.0.1版本的预测库* ）。
* 下载之后使用下面的方法解压。
......@@ -89,10 +89,11 @@ tar -xf paddle_inference.tgz
#### 1.2.2 预测库源码编译
* 如果希望获取最新预测库特性，可以从Paddle github上克隆最新代码，源码编译预测库。
* 可以参考[Paddle预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html)的说明，从github上获取Paddle代码，然后进行编译，生成最新的预测库。使用git获取代码方法如下。
* 可以参考[Paddle预测库安装编译说明](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi) 的说明，从github上获取Paddle代码，然后进行编译，生成最新的预测库。使用git获取代码方法如下。
```shell
git clone https://github.com/PaddlePaddle/Paddle.git
git checkout release/2.1
```
* 进入Paddle目录后，编译方法如下。
......@@ -115,7 +116,7 @@ make -j
make inference_lib_dist
```
更多编译参数选项可以参考Paddle C++预测库官网：[https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html)。
更多编译参数选项介绍可以参考[文档说明](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi)。
* 编译完成之后，可以在`build/paddle_inference_install_dir/`文件下看到生成了以下文件及文件夹。
......@@ -140,11 +141,11 @@ build/paddle_inference_install_dir/
```
inference/
|-- det_db
| |--inference.pdparams
| |--inference.pdimodel
| |--inference.pdiparams
| |--inference.pdmodel
|-- rec_rcnn
| |--inference.pdparams
| |--inference.pdparams
| |--inference.pdiparams
| |--inference.pdmodel
```
......
deploy/cpp_infer/readme_en.md
浏览文件 @ 253b8453
......@@ -78,8 +78,7 @@ opencv3/
#### 1.2.1 Direct download and installation
* Different cuda versions of the Linux inference library (based on GCC 4.8.2) are provided on the
[Paddle inference library official website](https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html). You can view and select the appropriate version of the inference library on the official website.
[Paddle inference library official website](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html). You can view and select the appropriate version of the inference library on the official website.
* After downloading, use the following method to uncompress.
......@@ -97,9 +96,10 @@ Finally you can see the following files in the folder of `paddle_inference/`.
```shell
git clone https://github.com/PaddlePaddle/Paddle.git
git checkout release/2.1
```
* After entering the Paddle directory, the compilation method is as follows.
* After entering the Paddle directory, the commands to compile the paddle inference library are as follows.
```shell
rm -rf build
......@@ -119,7 +119,7 @@ make -j
make inference_lib_dist
```
For more compilation parameter options, please refer to the official website of the Paddle C++ inference library:[https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html](https://www.paddlepaddle.org.cn/documentation/docs/en/develop/guides/05_inference_deployment/inference/build_and_install_lib_en.html).
For more compilation parameter options, please refer to the [document](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi).
* After the compilation process, you can see the following files in the folder of `build/paddle_inference_install_dir/`.
......@@ -144,11 +144,11 @@ Among them, `paddle` is the Paddle library required for C++ prediction later, an
```
inference/
|-- det_db
| |--inference.pdparams
| |--inference.pdimodel
| |--inference.pdiparams
| |--inference.pdmodel
|-- rec_rcnn
| |--inference.pdparams
| |--inference.pdparams
| |--inference.pdiparams
| |--inference.pdmodel
```
......
deploy/cpp_infer/src/main.cpp
浏览文件 @ 253b8453
......@@ -27,9 +27,12 @@
#include <fstream>
#include <numeric>
#include <glog/logging.h>
#include <include/config.h>
#include <include/ocr_det.h>
#include <include/ocr_rec.h>
#include <include/utility.h>
#include <sys/stat.h>
using namespace std;
using namespace cv;
......@@ -47,13 +50,8 @@ int main(int argc, char **argv) {
config.PrintConfigInfo();
std::string img_path(argv[2]);
cv::Mat srcimg = cv::imread(img_path, cv::IMREAD_COLOR);
if (!srcimg.data) {
std::cerr << "[ERROR] image read failed! image path: " << img_path << "\n";
exit(1);
}
std::vector<std::string> all_img_names;
Utility::GetAllFiles((char *)img_path.c_str(), all_img_names);
DBDetector det(config.det_model_dir, config.use_gpu, config.gpu_id,
config.gpu_mem, config.cpu_math_library_num_threads,
......@@ -76,7 +74,18 @@ int main(int argc, char **argv) {
config.use_tensorrt, config.use_fp16);
auto start = std::chrono::system_clock::now();
for (auto img_dir : all_img_names) {
LOG(INFO) << "The predict img: " << img_dir;
cv::Mat srcimg = cv::imread(img_dir, cv::IMREAD_COLOR);
if (!srcimg.data) {
std::cerr << "[ERROR] image read failed! image path: " << img_path
<< "\n";
exit(1);
}
std::vector<std::vector<std::vector<int>>> boxes;
det.Run(srcimg, boxes);
rec.Run(boxes, srcimg, cls);
......@@ -88,6 +97,7 @@ int main(int argc, char **argv) {
std::chrono::microseconds::period::num /
std::chrono::microseconds::period::den
<< "s" << std::endl;
}
return 0;
}
deploy/cpp_infer/src/ocr_det.cpp
浏览文件 @ 253b8453
......@@ -30,6 +30,42 @@ void DBDetector::LoadModel(const std::string &model_dir) {
this->use_fp16_ ? paddle_infer::Config::Precision::kHalf
: paddle_infer::Config::Precision::kFloat32,
false, false);
std::map<std::string, std::vector<int>> min_input_shape = {
{"x", {1, 3, 50, 50}},
{"conv2d_92.tmp_0", {1, 96, 20, 20}},
{"conv2d_91.tmp_0", {1, 96, 10, 10}},
{"nearest_interp_v2_1.tmp_0", {1, 96, 10, 10}},
{"nearest_interp_v2_2.tmp_0", {1, 96, 20, 20}},
{"nearest_interp_v2_3.tmp_0", {1, 24, 20, 20}},
{"nearest_interp_v2_4.tmp_0", {1, 24, 20, 20}},
{"nearest_interp_v2_5.tmp_0", {1, 24, 20, 20}},
{"elementwise_add_7", {1, 56, 2, 2}},
{"nearest_interp_v2_0.tmp_0", {1, 96, 2, 2}}};
std::map<std::string, std::vector<int>> max_input_shape = {
{"x", {1, 3, this->max_side_len_, this->max_side_len_}},
{"conv2d_92.tmp_0", {1, 96, 400, 400}},
{"conv2d_91.tmp_0", {1, 96, 200, 200}},
{"nearest_interp_v2_1.tmp_0", {1, 96, 200, 200}},
{"nearest_interp_v2_2.tmp_0", {1, 96, 400, 400}},
{"nearest_interp_v2_3.tmp_0", {1, 24, 400, 400}},
{"nearest_interp_v2_4.tmp_0", {1, 24, 400, 400}},
{"nearest_interp_v2_5.tmp_0", {1, 24, 400, 400}},
{"elementwise_add_7", {1, 56, 400, 400}},
{"nearest_interp_v2_0.tmp_0", {1, 96, 400, 400}}};
std::map<std::string, std::vector<int>> opt_input_shape = {
{"x", {1, 3, 640, 640}},
{"conv2d_92.tmp_0", {1, 96, 160, 160}},
{"conv2d_91.tmp_0", {1, 96, 80, 80}},
{"nearest_interp_v2_1.tmp_0", {1, 96, 80, 80}},
{"nearest_interp_v2_2.tmp_0", {1, 96, 160, 160}},
{"nearest_interp_v2_3.tmp_0", {1, 24, 160, 160}},
{"nearest_interp_v2_4.tmp_0", {1, 24, 160, 160}},
{"nearest_interp_v2_5.tmp_0", {1, 24, 160, 160}},
{"elementwise_add_7", {1, 56, 40, 40}},
{"nearest_interp_v2_0.tmp_0", {1, 96, 40, 40}}};
config.SetTRTDynamicShapeInfo(min_input_shape, max_input_shape,
opt_input_shape);
}
} else {
config.DisableGpu();
......@@ -48,7 +84,7 @@ void DBDetector::LoadModel(const std::string &model_dir) {
config.SwitchIrOptim(true);
config.EnableMemoryOptim();
config.DisableGlogInfo();
// config.DisableGlogInfo();
this->predictor_ = CreatePredictor(config);
}
......
deploy/cpp_infer/src/ocr_rec.cpp
浏览文件 @ 253b8453
......@@ -106,6 +106,15 @@ void CRNNRecognizer::LoadModel(const std::string &model_dir) {
this->use_fp16_ ? paddle_infer::Config::Precision::kHalf
: paddle_infer::Config::Precision::kFloat32,
false, false);
std::map<std::string, std::vector<int>> min_input_shape = {
{"x", {1, 3, 32, 10}}};
std::map<std::string, std::vector<int>> max_input_shape = {
{"x", {1, 3, 32, 2000}}};
std::map<std::string, std::vector<int>> opt_input_shape = {
{"x", {1, 3, 32, 320}}};
config.SetTRTDynamicShapeInfo(min_input_shape, max_input_shape,
opt_input_shape);
}
} else {
config.DisableGpu();
......
deploy/cpp_infer/src/preprocess_op.cpp 100755 → 100644
浏览文件 @ 253b8453
......@@ -47,16 +47,13 @@ void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
e /= 255.0;
}
(*im).convertTo(*im, CV_32FC3, e);
for (int h = 0; h < im->rows; h++) {
for (int w = 0; w < im->cols; w++) {
im->at<cv::Vec3f>(h, w)[0] =
(im->at<cv::Vec3f>(h, w)[0] - mean[0]) * scale[0];
im->at<cv::Vec3f>(h, w)[1] =
(im->at<cv::Vec3f>(h, w)[1] - mean[1]) * scale[1];
im->at<cv::Vec3f>(h, w)[2] =
(im->at<cv::Vec3f>(h, w)[2] - mean[2]) * scale[2];
}
std::vector<cv::Mat> bgr_channels(3);
cv::split(*im, bgr_channels);
for (auto i = 0; i < bgr_channels.size(); i++) {
bgr_channels[i].convertTo(bgr_channels[i], CV_32FC1, 1.0 * scale[i],
(0.0 - mean[i]) * scale[i]);
}
cv::merge(bgr_channels, *im);
}
void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
......@@ -81,15 +78,9 @@ void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
resize_h = max(int(round(float(resize_h) / 32) * 32), 32);
resize_w = max(int(round(float(resize_w) / 32) * 32), 32);
if (!use_tensorrt) {
cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
ratio_h = float(resize_h) / float(h);
ratio_w = float(resize_w) / float(w);
} else {
cv::resize(img, resize_img, cv::Size(640, 640));
ratio_h = float(640) / float(h);
ratio_w = float(640) / float(w);
}
}
void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
......@@ -108,23 +99,12 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
resize_w = imgW;
else
resize_w = int(ceilf(imgH * ratio));
if (!use_tensorrt) {
cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
cv::INTER_LINEAR);
cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
{127, 127, 127});
} else {
int k = int(img.cols * 32 / img.rows);
if (k >= 100) {
cv::resize(img, resize_img, cv::Size(100, 32), 0.f, 0.f,
cv::INTER_LINEAR);
} else {
cv::resize(img, resize_img, cv::Size(k, 32), 0.f, 0.f, cv::INTER_LINEAR);
cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, int(100 - k),
cv::BORDER_CONSTANT, {127, 127, 127});
}
}
}
void ClsResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
......@@ -142,16 +122,12 @@ void ClsResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
else
resize_w = int(ceilf(imgH * ratio));
if (!use_tensorrt) {
cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
cv::INTER_LINEAR);
if (resize_w < imgW) {
cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, imgW - resize_w,
cv::BORDER_CONSTANT, cv::Scalar(0, 0, 0));
}
} else {
cv::resize(img, resize_img, cv::Size(100, 32), 0.f, 0.f, cv::INTER_LINEAR);
}
}
} // namespace PaddleOCR
deploy/cpp_infer/src/utility.cpp
浏览文件 @ 253b8453
......@@ -12,12 +12,14 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include <dirent.h>
#include <include/utility.h>
#include <iostream>
#include <ostream>
#include <sys/stat.h>
#include <sys/types.h>
#include <vector>
#include <include/utility.h>
namespace PaddleOCR {
std::vector<std::string> Utility::ReadDict(const std::string &path) {
......@@ -57,4 +59,37 @@ void Utility::VisualizeBboxes(
<< std::endl;
}
// list all files under a directory
void Utility::GetAllFiles(const char *dir_name,
std::vector<std::string> &all_inputs) {
if (NULL == dir_name) {
std::cout << " dir_name is null ! " << std::endl;
return;
}
struct stat s;
lstat(dir_name, &s);
if (!S_ISDIR(s.st_mode)) {
std::cout << "dir_name is not a valid directory !" << std::endl;
all_inputs.push_back(dir_name);
return;
} else {
struct dirent *filename; // return value for readdir()
DIR *dir; // return value for opendir()
dir = opendir(dir_name);
if (NULL == dir) {
std::cout << "Can not open dir " << dir_name << std::endl;
return;
}
std::cout << "Successfully opened the dir !" << std::endl;
while ((filename = readdir(dir)) != NULL) {
if (strcmp(filename->d_name, ".") == 0 ||
strcmp(filename->d_name, "..") == 0)
continue;
// img_dir + std::string("/") + all_inputs[0];
all_inputs.push_back(dir_name + std::string("/") +
std::string(filename->d_name));
}
}
}
} // namespace PaddleOCR
\ No newline at end of file
deploy/cpp_infer/tools/build.sh
浏览文件 @ 253b8453
......@@ -12,9 +12,10 @@ cmake .. \
-DWITH_MKL=ON \
-DWITH_GPU=OFF \
-DWITH_STATIC_LIB=OFF \
-DUSE_TENSORRT=OFF \
-DWITH_TENSORRT=OFF \
-DOPENCV_DIR=${OPENCV_DIR} \
-DCUDNN_LIB=${CUDNN_LIB_DIR} \
-DCUDA_LIB=${CUDA_LIB_DIR} \
-DTENSORRT_DIR=${TENSORRT_DIR} \
make -j
deploy/cpp_infer/tools/config.txt
浏览文件 @ 253b8453
......@@ -23,7 +23,7 @@ rec_model_dir ./inference/ch_ppocr_mobile_v2.0_rec_infer/
char_list_file ../../ppocr/utils/ppocr_keys_v1.txt
# show the detection results
visualize 1
visualize 0
# use_tensorrt
use_tensorrt 0
......
deploy/hubserving/readme.md
浏览文件 @ 253b8453
......@@ -29,7 +29,7 @@ deploy/hubserving/ocr_system/
### 1. 准备环境
```shell
# 安装paddlehub
pip3 install paddlehub --upgrade -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install paddlehub==1.8.3 --upgrade -i https://pypi.tuna.tsinghua.edu.cn/simple
```
### 2. 下载推理模型
......
deploy/hubserving/readme_en.md
浏览文件 @ 253b8453
......@@ -30,7 +30,7 @@ The following steps take the 2-stage series service as an example. If only the d
### 1. Prepare the environment
```shell
# Install paddlehub
pip3 install paddlehub --upgrade -i https://pypi.tuna.tsinghua.edu.cn/simple
pip3 install paddlehub==1.8.3 --upgrade -i https://pypi.tuna.tsinghua.edu.cn/simple
```
### 2. Download inference model
......
doc/doc_ch/config.md
浏览文件 @ 253b8453
......@@ -111,9 +111,9 @@
| 字段 | 用途 | 默认值 | 备注 |
| :---------------------: | :---------------------: | :--------------: | :--------------------: |
| **dataset** | 每次迭代返回一个样本 | - | - |
| name | dataset类名 | SimpleDataSet | 目前支持`SimpleDataSet`和`LMDBDateSet` |
| name | dataset类名 | SimpleDataSet | 目前支持`SimpleDataSet`和`LMDBDataSet` |
| data_dir | 数据集图片存放路径 | ./train_data | \ |
| label_file_list | 数据标签路径 | ["./train_data/train_list.txt"] | dataset为LMDBDateSet时不需要此参数 |
| label_file_list | 数据标签路径 | ["./train_data/train_list.txt"] | dataset为LMDBDataSet时不需要此参数 |
| ratio_list | 数据集的比例 | [1.0] | 若label_file_list中有两个train_list，且ratio_list为[0.4,0.6]，则从train_list1中采样40%，从train_list2中采样60%组合整个dataset |
| transforms | 对图片和标签进行变换的方法列表 | [DecodeImage,CTCLabelEncode,RecResizeImg,KeepKeys] | 见[ppocr/data/imaug](../../ppocr/data/imaug) |
| **loader** | dataloader相关 | - | |
......
doc/doc_ch/recognition.md
浏览文件 @ 253b8453
......@@ -243,7 +243,7 @@ Optimizer:
Train:
dataset:
# 数据集格式，支持LMDBDateSet以及SimpleDataSet
# 数据集格式，支持LMDBDataSet以及SimpleDataSet
name: SimpleDataSet
# 数据集路径
data_dir: ./train_data/
......@@ -263,7 +263,7 @@ Train:
Eval:
dataset:
# 数据集格式，支持LMDBDateSet以及SimpleDataSet
# 数据集格式，支持LMDBDataSet以及SimpleDataSet
name: SimpleDataSet
# 数据集路径
data_dir: ./train_data
......@@ -393,7 +393,7 @@ Global:
Train:
dataset:
# 数据集格式，支持LMDBDateSet以及SimpleDataSet
# 数据集格式，支持LMDBDataSet以及SimpleDataSet
name: SimpleDataSet
# 数据集路径
data_dir: ./train_data/
......@@ -403,7 +403,7 @@ Train:
Eval:
dataset:
# 数据集格式，支持LMDBDateSet以及SimpleDataSet
# 数据集格式，支持LMDBDataSet以及SimpleDataSet
name: SimpleDataSet
# 数据集路径
data_dir: ./train_data
......
doc/doc_ch/whl.md
浏览文件 @ 253b8453
......@@ -59,7 +59,7 @@ im_show.save('result.jpg')
from paddleocr import PaddleOCR, draw_ocr
ocr = PaddleOCR() # need to run only once to download and load model into memory
img_path = 'PaddleOCR/doc/imgs/11.jpg'
result = ocr.ocr(img_path)
result = ocr.ocr(img_path,cls=False)
for line in result:
print(line)
......@@ -355,3 +355,4 @@ im_show.save('result.jpg')
| det | 前向时使用启动检测 | TRUE |
| rec | 前向时是否启动识别 | TRUE |
| cls | 前向时是否启动分类 (命令行模式下使用use_angle_cls控制前向是否启动分类) | FALSE |
| show_log | 是否打印det和rec等信息 | FALSE |
doc/doc_en/config_en.md
浏览文件 @ 253b8453
......@@ -110,9 +110,9 @@ In ppocr, the network is divided into four stages: Transform, Backbone, Neck and
| Parameter | Use | Defaults | Note |
| :---------------------: | :---------------------: | :--------------: | :--------------------: |
| **dataset** | Return one sample per iteration | - | - |
| name | dataset class name | SimpleDataSet | Currently support`SimpleDataSet`,`LMDBDateSet` |
| name | dataset class name | SimpleDataSet | Currently support`SimpleDataSet`,`LMDBDataSet` |
| data_dir | Image folder path | ./train_data | \ |
| label_file_list | Groundtruth file path | ["./train_data/train_list.txt"] | This parameter is not required when dataset is LMDBDateSet |
| label_file_list | Groundtruth file path | ["./train_data/train_list.txt"] | This parameter is not required when dataset is LMDBDataSet |
| ratio_list | Ratio of data set | [1.0] | If there are two train_lists in label_file_list and ratio_list is [0.4,0.6], 40% will be sampled from train_list1, and 60% will be sampled from train_list2 to combine the entire dataset |
| transforms | List of methods to transform images and labels | [DecodeImage,CTCLabelEncode,RecResizeImg,KeepKeys] | see[ppocr/data/imaug](../../ppocr/data/imaug) |
| **loader** | dataloader related | - | |
......
doc/doc_en/distributed_training.md 0 → 100644
浏览文件 @ 253b8453
# Distributed training
## Introduction
The high performance of distributed training is one of the core advantages of PaddlePaddle. In the classification task, distributed training can achieve almost linear speedup ratio. Generally, OCR training task need massive training data. Such as recognition, ppocrv2.0 model is trained based on 1800W dataset, which is very time-consuming if using single machine. Therefore, the distributed training is used in paddleocr to speedup the training task. For more information about distributed training, please refer to [distributed training quick start tutorial](https://fleet-x.readthedocs.io/en/latest/paddle_fleet_rst/parameter_server/ps_quick_start.html).
## Quick Start
### Training with single machine
Take recognition as an example. After the data is prepared locally, start the training task with the interface of `paddle.distributed.launch`. The start command as follows:
```shell
python3 -m paddle.distributed.launch \
--log_dir=./log/ \
--gpus '0,1,2,3,4,5,6,7' \
tools/train.py \
-c configs/rec/rec_mv3_none_bilstm_ctc.yml
```
### Training with multi machine
Compared with single machine, training with multi machine only needs to add the parameter `--ips` to start command, which represents the IP list of machines used for distributed training, and the IP of different machines are separated by commas. The start command as follows:
```shell
ip_list="192.168.0.1,192.168.0.2"
python3 -m paddle.distributed.launch \
--log_dir=./log/ \
--ips="${ip_list}" \
--gpus="0,1,2,3,4,5,6,7" \
tools/train.py \
-c configs/rec/rec_mv3_none_bilstm_ctc.yml
```
**Notice:**
* The IP addresses of different machines need to be separated by commas, which can be queried through `ifconfig` or `ipconfig`.
* Different machines need to be set to be secret free and can `ping` success with others directly, otherwise communication cannot establish between them.
* The code, data and start command betweent different machines must be completely consistent and then all machines need to run start command. The first machine in the `ip_list` is set to `trainer0`, and so on.
## Performance comparison
* Based on 26W public recognition dataset (LSVT, rctw, mtwi), training on single 8-card P40 and dual 8-card P40, the final time consumption is as follows.
| Model | Config file | Number of machines | Number of GPUs per machine | Training time | Recognition acc | Speedup ratio |
| :-------: | :------------: | :----------------: | :----------------------------: | :------------------: | :--------------: | :-----------: |
| CRNN | configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml | 1 | 8 | 60h | 66.7% | - |
| CRNN | configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml | 2 | 8 | 40h | 67.0% | 150% |
It can be seen that the training time is shortened from 60h to 40h, the speedup ratio can reach 150% (60h / 40h), and the efficiency is 75% (60h / (40h * 2)).
doc/doc_en/recognition_en.md
浏览文件 @ 253b8453
......@@ -237,7 +237,7 @@ Optimizer:
Train:
dataset:
# Type of dataset，we support LMDBDateSet and SimpleDataSet
# Type of dataset，we support LMDBDataSet and SimpleDataSet
name: SimpleDataSet
# Path of dataset
data_dir: ./train_data/
......@@ -257,7 +257,7 @@ Train:
Eval:
dataset:
# Type of dataset，we support LMDBDateSet and SimpleDataSet
# Type of dataset，we support LMDBDataSet and SimpleDataSet
name: SimpleDataSet
# Path of dataset
data_dir: ./train_data
......@@ -394,7 +394,7 @@ Global:
Train:
dataset:
# Type of dataset，we support LMDBDateSet and SimpleDataSet
# Type of dataset，we support LMDBDataSet and SimpleDataSet
name: SimpleDataSet
# Path of dataset
data_dir: ./train_data/
......@@ -404,7 +404,7 @@ Train:
Eval:
dataset:
# Type of dataset，we support LMDBDateSet and SimpleDataSet
# Type of dataset，we support LMDBDataSet and SimpleDataSet
name: SimpleDataSet
# Path of dataset
data_dir: ./train_data
......
doc/doc_en/whl_en.md
浏览文件 @ 253b8453
......@@ -59,7 +59,7 @@ Visualization of results
from paddleocr import PaddleOCR,draw_ocr
ocr = PaddleOCR(lang='en') # need to run only once to download and load model into memory
img_path = 'PaddleOCR/doc/imgs_en/img_12.jpg'
result = ocr.ocr(img_path)
result = ocr.ocr(img_path, cls=False)
for line in result:
print(line)
......@@ -362,3 +362,5 @@ im_show.save('result.jpg')
| det | Enable detction when `ppocr.ocr` func exec | TRUE |
| rec | Enable recognition when `ppocr.ocr` func exec | TRUE |
| cls | Enable classification when `ppocr.ocr` func exec((Use use_angle_cls in command line mode to control whether to start classification in the forward direction) | FALSE |
| show_log | Whether to print log in det and rec
| FALSE |
\ No newline at end of file
paddleocr.py
浏览文件 @ 253b8453
......@@ -19,18 +19,17 @@ __dir__ = os.path.dirname(__file__)
sys.path.append(os.path.join(__dir__, ''))
import cv2
import logging
import numpy as np
from pathlib import Path
import tarfile
import requests
from tqdm import tqdm
from tools.infer import predict_system
from ppocr.utils.logging import get_logger
logger = get_logger()
from ppocr.utils.utility import check_and_read_gif, get_image_file_list
from tools.infer.utility import draw_ocr
from ppocr.utils.network import maybe_download, download_with_progressbar, is_link, confirm_model_dir_url
from tools.infer.utility import draw_ocr, init_args, str2bool
__all__ = ['PaddleOCR']
......@@ -123,150 +122,24 @@ SUPPORT_REC_MODEL = ['CRNN']
BASE_DIR = os.path.expanduser("~/.paddleocr/")
def download_with_progressbar(url, save_path):
response = requests.get(url, stream=True)
total_size_in_bytes = int(response.headers.get('content-length', 0))
block_size = 1024 # 1 Kibibyte
progress_bar = tqdm(total=total_size_in_bytes, unit='iB', unit_scale=True)
with open(save_path, 'wb') as file:
for data in response.iter_content(block_size):
progress_bar.update(len(data))
file.write(data)
progress_bar.close()
if total_size_in_bytes == 0 or progress_bar.n != total_size_in_bytes:
logger.error("Something went wrong while downloading models")
sys.exit(0)
def maybe_download(model_storage_directory, url):
# using custom model
tar_file_name_list = [
'inference.pdiparams', 'inference.pdiparams.info', 'inference.pdmodel'
]
if not os.path.exists(
os.path.join(model_storage_directory, 'inference.pdiparams')
) or not os.path.exists(
os.path.join(model_storage_directory, 'inference.pdmodel')):
tmp_path = os.path.join(model_storage_directory, url.split('/')[-1])
print('download {} to {}'.format(url, tmp_path))
os.makedirs(model_storage_directory, exist_ok=True)
download_with_progressbar(url, tmp_path)
with tarfile.open(tmp_path, 'r') as tarObj:
for member in tarObj.getmembers():
filename = None
for tar_file_name in tar_file_name_list:
if tar_file_name in member.name:
filename = tar_file_name
if filename is None:
continue
file = tarObj.extractfile(member)
with open(
os.path.join(model_storage_directory, filename),
'wb') as f:
f.write(file.read())
os.remove(tmp_path)
def parse_args(mMain=True, add_help=True):
def parse_args(mMain=True):
import argparse
def str2bool(v):
return v.lower() in ("true", "t", "1")
if mMain:
parser = argparse.ArgumentParser(add_help=add_help)
# params for prediction engine
parser.add_argument("--use_gpu", type=str2bool, default=True)
parser.add_argument("--ir_optim", type=str2bool, default=True)
parser.add_argument("--use_tensorrt", type=str2bool, default=False)
parser.add_argument("--gpu_mem", type=int, default=8000)
# params for text detector
parser.add_argument("--image_dir", type=str)
parser.add_argument("--det_algorithm", type=str, default='DB')
parser.add_argument("--det_model_dir", type=str, default=None)
parser.add_argument("--det_limit_side_len", type=float, default=960)
parser.add_argument("--det_limit_type", type=str, default='max')
# DB parmas
parser.add_argument("--det_db_thresh", type=float, default=0.3)
parser.add_argument("--det_db_box_thresh", type=float, default=0.5)
parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6)
parser.add_argument("--use_dilation", type=bool, default=False)
parser.add_argument("--det_db_score_mode", type=str, default="fast")
# EAST parmas
parser.add_argument("--det_east_score_thresh", type=float, default=0.8)
parser.add_argument("--det_east_cover_thresh", type=float, default=0.1)
parser.add_argument("--det_east_nms_thresh", type=float, default=0.2)
# params for text recognizer
parser.add_argument("--rec_algorithm", type=str, default='CRNN')
parser.add_argument("--rec_model_dir", type=str, default=None)
parser.add_argument("--rec_image_shape", type=str, default="3, 32, 320")
parser.add_argument("--rec_char_type", type=str, default='ch')
parser.add_argument("--rec_batch_num", type=int, default=6)
parser.add_argument("--max_text_length", type=int, default=25)
parser.add_argument("--rec_char_dict_path", type=str, default=None)
parser.add_argument("--use_space_char", type=bool, default=True)
parser.add_argument("--drop_score", type=float, default=0.5)
# params for text classifier
parser.add_argument("--cls_model_dir", type=str, default=None)
parser.add_argument("--cls_image_shape", type=str, default="3, 48, 192")
parser.add_argument("--label_list", type=list, default=['0', '180'])
parser.add_argument("--cls_batch_num", type=int, default=6)
parser.add_argument("--cls_thresh", type=float, default=0.9)
parser.add_argument("--enable_mkldnn", type=bool, default=False)
parser.add_argument("--use_zero_copy_run", type=bool, default=False)
parser.add_argument("--use_pdserving", type=str2bool, default=False)
parser = init_args()
parser.add_help = mMain
parser.add_argument("--lang", type=str, default='ch')
parser.add_argument("--det", type=str2bool, default=True)
parser.add_argument("--rec", type=str2bool, default=True)
parser.add_argument("--use_angle_cls", type=str2bool, default=False)
for action in parser._actions:
if action.dest == 'rec_char_dict_path':
action.default = None
if mMain:
return parser.parse_args()
else:
return argparse.Namespace(
use_gpu=True,
ir_optim=True,
use_tensorrt=False,
gpu_mem=8000,
image_dir='',
det_algorithm='DB',
det_model_dir=None,
det_limit_side_len=960,
det_limit_type='max',
det_db_thresh=0.3,
det_db_box_thresh=0.5,
det_db_unclip_ratio=1.6,
use_dilation=False,
det_db_score_mode="fast",
det_east_score_thresh=0.8,
det_east_cover_thresh=0.1,
det_east_nms_thresh=0.2,
rec_algorithm='CRNN',
rec_model_dir=None,
rec_image_shape="3, 32, 320",
rec_char_type='ch',
rec_batch_num=6,
max_text_length=25,
rec_char_dict_path=None,
use_space_char=True,
drop_score=0.5,
cls_model_dir=None,
cls_image_shape="3, 48, 192",
label_list=['0', '180'],
cls_batch_num=6,
cls_thresh=0.9,
enable_mkldnn=False,
use_zero_copy_run=False,
use_pdserving=False,
lang='ch',
det=True,
rec=True,
use_angle_cls=False)
inference_args_dict = {}
for action in parser._actions:
inference_args_dict[action.dest] = action.default
return argparse.Namespace(**inference_args_dict)
class PaddleOCR(predict_system.TextSystem):
......@@ -276,10 +149,12 @@ class PaddleOCR(predict_system.TextSystem):
args:
**kwargs: other params show in paddleocr --help
"""
postprocess_params = parse_args(mMain=False, add_help=False)
postprocess_params.__dict__.update(**kwargs)
self.use_angle_cls = postprocess_params.use_angle_cls
lang = postprocess_params.lang
params = parse_args(mMain=False)
params.__dict__.update(**kwargs)
if not params.show_log:
logger.setLevel(logging.INFO)
self.use_angle_cls = params.use_angle_cls
lang = params.lang
latin_lang = [
'af', 'az', 'bs', 'cs', 'cy', 'da', 'de', 'es', 'et', 'fr', 'ga',
'hr', 'hu', 'id', 'is', 'it', 'ku', 'la', 'lt', 'lv', 'mi', 'ms',
......@@ -311,42 +186,41 @@ class PaddleOCR(predict_system.TextSystem):
else:
det_lang = "en"
use_inner_dict = False
if postprocess_params.rec_char_dict_path is None:
if params.rec_char_dict_path is None:
use_inner_dict = True
postprocess_params.rec_char_dict_path = model_urls['rec'][lang][
params.rec_char_dict_path = model_urls['rec'][lang][
'dict_path']
# init model dir
if postprocess_params.det_model_dir is None:
postprocess_params.det_model_dir = os.path.join(BASE_DIR, VERSION,
'det', det_lang)
if postprocess_params.rec_model_dir is None:
postprocess_params.rec_model_dir = os.path.join(BASE_DIR, VERSION,
'rec', lang)
if postprocess_params.cls_model_dir is None:
postprocess_params.cls_model_dir = os.path.join(BASE_DIR, 'cls')
print(postprocess_params)
# download model
maybe_download(postprocess_params.det_model_dir,
params.det_model_dir, det_url = confirm_model_dir_url(params.det_model_dir,
os.path.join(BASE_DIR, VERSION, 'det', det_lang),
model_urls['det'][det_lang])
maybe_download(postprocess_params.rec_model_dir,
params.rec_model_dir, rec_url = confirm_model_dir_url(params.rec_model_dir,
os.path.join(BASE_DIR, VERSION, 'rec', lang),
model_urls['rec'][lang]['url'])
maybe_download(postprocess_params.cls_model_dir, model_urls['cls'])
params.cls_model_dir, cls_url = confirm_model_dir_url(params.cls_model_dir,
os.path.join(BASE_DIR, VERSION, 'cls'),
model_urls['cls'])
# download model
maybe_download(params.det_model_dir, det_url)
maybe_download(params.rec_model_dir, rec_url)
maybe_download(params.cls_model_dir, cls_url)
if postprocess_params.det_algorithm not in SUPPORT_DET_MODEL:
if params.det_algorithm not in SUPPORT_DET_MODEL:
logger.error('det_algorithm must in {}'.format(SUPPORT_DET_MODEL))
sys.exit(0)
if postprocess_params.rec_algorithm not in SUPPORT_REC_MODEL:
if params.rec_algorithm not in SUPPORT_REC_MODEL:
logger.error('rec_algorithm must in {}'.format(SUPPORT_REC_MODEL))
sys.exit(0)
if use_inner_dict:
postprocess_params.rec_char_dict_path = str(
Path(__file__).parent / postprocess_params.rec_char_dict_path)
params.rec_char_dict_path = str(
Path(__file__).parent / params.rec_char_dict_path)
print(params)
# init det_model and rec_model
super().__init__(postprocess_params)
super().__init__(params)
def ocr(self, img, det=True, rec=True, cls=False):
def ocr(self, img, det=True, rec=True, cls=True):
"""
ocr with paddleocr
args：
......@@ -358,9 +232,7 @@ class PaddleOCR(predict_system.TextSystem):
if isinstance(img, list) and det == True:
logger.error('When input a list of images, det must be false')
exit(0)
if cls == False:
self.use_angle_cls = False
elif cls == True and self.use_angle_cls == False:
if cls == True and self.use_angle_cls == False:
logger.warning(
'Since the angle classifier is not initialized, the angle classifier will not be uesd during the forward process'
)
......@@ -382,7 +254,7 @@ class PaddleOCR(predict_system.TextSystem):
if isinstance(img, np.ndarray) and len(img.shape) == 2:
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
if det and rec:
dt_boxes, rec_res = self.__call__(img)
dt_boxes, rec_res = self.__call__(img, cls)
return [[box.tolist(), res] for box, res in zip(dt_boxes, rec_res)]
elif det and not rec:
dt_boxes, elapse = self.text_detector(img)
......@@ -392,7 +264,7 @@ class PaddleOCR(predict_system.TextSystem):
else:
if not isinstance(img, list):
img = [img]
if self.use_angle_cls:
if self.use_angle_cls and cls:
img, cls_res, elapse = self.text_classifier(img)
if not rec:
return cls_res
......@@ -404,7 +276,7 @@ def main():
# for cmd
args = parse_args(mMain=True)
image_dir = args.image_dir
if image_dir.startswith('http'):
if is_link(image_dir):
download_with_progressbar(image_dir, 'tmp.jpg')
image_file_list = ['tmp.jpg']
else:
......
ppocr/data/__init__.py
浏览文件 @ 253b8453
......@@ -35,6 +35,7 @@ from ppocr.data.imaug import transform, create_operators
from ppocr.data.simple_dataset import SimpleDataSet
from ppocr.data.lmdb_dataset import LMDBDataSet
from ppocr.data.pgnet_dataset import PGDataSet
from ppocr.data.pubtab_dataset import PubTabDataSet
__all__ = ['build_dataloader', 'transform', 'create_operators']
......@@ -55,7 +56,7 @@ signal.signal(signal.SIGTERM, term_mp)
def build_dataloader(config, mode, device, logger, seed=None):
config = copy.deepcopy(config)
support_dict = ['SimpleDataSet', 'LMDBDataSet', 'PGDataSet']
support_dict = ['SimpleDataSet', 'LMDBDataSet', 'PGDataSet', 'PubTabDataSet']
module_name = config[mode]['dataset']['name']
assert module_name in support_dict, Exception(
'DataSet only support {}'.format(support_dict))
......
ppocr/data/imaug/__init__.py
浏览文件 @ 253b8453
......@@ -29,6 +29,7 @@ from .label_ops import *
from .east_process import *
from .sast_process import *
from .pg_process import *
from .gen_table_mask import *
def transform(data, ops=None):
......
ppocr/data/imaug/gen_table_mask.py 0 → 100644
浏览文件 @ 253b8453
"""
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals
import sys
import six
import cv2
import numpy as np
class GenTableMask(object):
""" gen table mask """
def __init__(self, shrink_h_max, shrink_w_max, mask_type=0, **kwargs):
self.shrink_h_max = 5
self.shrink_w_max = 5
self.mask_type = mask_type
def projection(self, erosion, h, w, spilt_threshold=0):
# 水平投影
projection_map = np.ones_like(erosion)
project_val_array = [0 for _ in range(0, h)]
for j in range(0, h):
for i in range(0, w):
if erosion[j, i] == 255:
project_val_array[j] += 1
# 根据数组，获取切割点
start_idx = 0 # 记录进入字符区的索引
end_idx = 0 # 记录进入空白区域的索引
in_text = False # 是否遍历到了字符区内
box_list = []
for i in range(len(project_val_array)):
if in_text == False and project_val_array[i] > spilt_threshold: # 进入字符区了
in_text = True
start_idx = i
elif project_val_array[i] <= spilt_threshold and in_text == True: # 进入空白区了
end_idx = i
in_text = False
if end_idx - start_idx <= 2:
continue
box_list.append((start_idx, end_idx + 1))
if in_text:
box_list.append((start_idx, h - 1))
# 绘制投影直方图
for j in range(0, h):
for i in range(0, project_val_array[j]):
projection_map[j, i] = 0
return box_list, projection_map
def projection_cx(self, box_img):
box_gray_img = cv2.cvtColor(box_img, cv2.COLOR_BGR2GRAY)
h, w = box_gray_img.shape
# 灰度图片进行二值化处理
ret, thresh1 = cv2.threshold(box_gray_img, 200, 255, cv2.THRESH_BINARY_INV)
# 纵向腐蚀
if h < w:
kernel = np.ones((2, 1), np.uint8)
erode = cv2.erode(thresh1, kernel, iterations=1)
else:
erode = thresh1
# 水平膨胀
kernel = np.ones((1, 5), np.uint8)
erosion = cv2.dilate(erode, kernel, iterations=1)
# 水平投影
projection_map = np.ones_like(erosion)
project_val_array = [0 for _ in range(0, h)]
for j in range(0, h):
for i in range(0, w):
if erosion[j, i] == 255:
project_val_array[j] += 1
# 根据数组，获取切割点
start_idx = 0 # 记录进入字符区的索引
end_idx = 0 # 记录进入空白区域的索引
in_text = False # 是否遍历到了字符区内
box_list = []
spilt_threshold = 0
for i in range(len(project_val_array)):
if in_text == False and project_val_array[i] > spilt_threshold: # 进入字符区了
in_text = True
start_idx = i
elif project_val_array[i] <= spilt_threshold and in_text == True: # 进入空白区了
end_idx = i
in_text = False
if end_idx - start_idx <= 2:
continue
box_list.append((start_idx, end_idx + 1))
if in_text:
box_list.append((start_idx, h - 1))
# 绘制投影直方图
for j in range(0, h):
for i in range(0, project_val_array[j]):
projection_map[j, i] = 0
split_bbox_list = []
if len(box_list) > 1:
for i, (h_start, h_end) in enumerate(box_list):
if i == 0:
h_start = 0
if i == len(box_list):
h_end = h
word_img = erosion[h_start:h_end + 1, :]
word_h, word_w = word_img.shape
w_split_list, w_projection_map = self.projection(word_img.T, word_w, word_h)
w_start, w_end = w_split_list[0][0], w_split_list[-1][1]
if h_start > 0:
h_start -= 1
h_end += 1
word_img = box_img[h_start:h_end + 1:, w_start:w_end + 1, :]
split_bbox_list.append([w_start, h_start, w_end, h_end])
else:
split_bbox_list.append([0, 0, w, h])
return split_bbox_list
def shrink_bbox(self, bbox):
left, top, right, bottom = bbox
sh_h = min(max(int((bottom - top) * 0.1), 1), self.shrink_h_max)
sh_w = min(max(int((right - left) * 0.1), 1), self.shrink_w_max)
left_new = left + sh_w
right_new = right - sh_w
top_new = top + sh_h
bottom_new = bottom - sh_h
if left_new >= right_new:
left_new = left
right_new = right
if top_new >= bottom_new:
top_new = top
bottom_new = bottom
return [left_new, top_new, right_new, bottom_new]
def __call__(self, data):
img = data['image']
cells = data['cells']
height, width = img.shape[0:2]
if self.mask_type == 1:
mask_img = np.zeros((height, width), dtype=np.float32)
else:
mask_img = np.zeros((height, width, 3), dtype=np.float32)
cell_num = len(cells)
for cno in range(cell_num):
if "bbox" in cells[cno]:
bbox = cells[cno]['bbox']
left, top, right, bottom = bbox
box_img = img[top:bottom, left:right, :].copy()
split_bbox_list = self.projection_cx(box_img)
for sno in range(len(split_bbox_list)):
split_bbox_list[sno][0] += left
split_bbox_list[sno][1] += top
split_bbox_list[sno][2] += left
split_bbox_list[sno][3] += top
for sno in range(len(split_bbox_list)):
left, top, right, bottom = split_bbox_list[sno]
left, top, right, bottom = self.shrink_bbox([left, top, right, bottom])
if self.mask_type == 1:
mask_img[top:bottom, left:right] = 1.0
data['mask_img'] = mask_img
else:
mask_img[top:bottom, left:right, :] = (255, 255, 255)
data['image'] = mask_img
return data
class ResizeTableImage(object):
def __init__(self, max_len, **kwargs):
super(ResizeTableImage, self).__init__()
self.max_len = max_len
def get_img_bbox(self, cells):
bbox_list = []
if len(cells) == 0:
return bbox_list
cell_num = len(cells)
for cno in range(cell_num):
if "bbox" in cells[cno]:
bbox = cells[cno]['bbox']
bbox_list.append(bbox)
return bbox_list
def resize_img_table(self, img, bbox_list, max_len):
height, width = img.shape[0:2]
ratio = max_len / (max(height, width) * 1.0)
resize_h = int(height * ratio)
resize_w = int(width * ratio)
img_new = cv2.resize(img, (resize_w, resize_h))
bbox_list_new = []
for bno in range(len(bbox_list)):
left, top, right, bottom = bbox_list[bno].copy()
left = int(left * ratio)
top = int(top * ratio)
right = int(right * ratio)
bottom = int(bottom * ratio)
bbox_list_new.append([left, top, right, bottom])
return img_new, bbox_list_new
def __call__(self, data):
img = data['image']
if 'cells' not in data:
cells = []
else:
cells = data['cells']
bbox_list = self.get_img_bbox(cells)
img_new, bbox_list_new = self.resize_img_table(img, bbox_list, self.max_len)
data['image'] = img_new
cell_num = len(cells)
bno = 0
for cno in range(cell_num):
if "bbox" in data['cells'][cno]:
data['cells'][cno]['bbox'] = bbox_list_new[bno]
bno += 1
data['max_len'] = self.max_len
return data
class PaddingTableImage(object):
def __init__(self, **kwargs):
super(PaddingTableImage, self).__init__()
def __call__(self, data):
img = data['image']
max_len = data['max_len']
padding_img = np.zeros((max_len, max_len, 3), dtype=np.float32)
height, width = img.shape[0:2]
padding_img[0:height, 0:width, :] = img.copy()
data['image'] = padding_img
return data
\ No newline at end of file
ppocr/data/imaug/label_ops.py
浏览文件 @ 253b8453
......@@ -351,3 +351,162 @@ class SRNLabelEncode(BaseRecLabelEncode):
assert False, "Unsupport type %s in get_beg_end_flag_idx" \
% beg_or_end
return idx
class TableLabelEncode(object):
""" Convert between text-label and text-index """
def __init__(self,
max_text_length,
max_elem_length,
max_cell_num,
character_dict_path,
span_weight = 1.0,
**kwargs):
self.max_text_length = max_text_length
self.max_elem_length = max_elem_length
self.max_cell_num = max_cell_num
list_character, list_elem = self.load_char_elem_dict(character_dict_path)
list_character = self.add_special_char(list_character)
list_elem = self.add_special_char(list_elem)
self.dict_character = {}
for i, char in enumerate(list_character):
self.dict_character[char] = i
self.dict_elem = {}
for i, elem in enumerate(list_elem):
self.dict_elem[elem] = i
self.span_weight = span_weight
def load_char_elem_dict(self, character_dict_path):
list_character = []
list_elem = []
with open(character_dict_path, "rb") as fin:
lines = fin.readlines()
substr = lines[0].decode('utf-8').strip("\n").split("\t")
character_num = int(substr[0])
elem_num = int(substr[1])
for cno in range(1, 1+character_num):
character = lines[cno].decode('utf-8').strip("\n")
list_character.append(character)
for eno in range(1+character_num, 1+character_num+elem_num):
elem = lines[eno].decode('utf-8').strip("\n")
list_elem.append(elem)
return list_character, list_elem
def add_special_char(self, list_character):
self.beg_str = "sos"
self.end_str = "eos"
list_character = [self.beg_str] + list_character + [self.end_str]
return list_character
def get_span_idx_list(self):
span_idx_list = []
for elem in self.dict_elem:
if 'span' in elem:
span_idx_list.append(self.dict_elem[elem])
return span_idx_list
def __call__(self, data):
cells = data['cells']
structure = data['structure']['tokens']
structure = self.encode(structure, 'elem')
if structure is None:
return None
elem_num = len(structure)
structure = [0] + structure + [len(self.dict_elem) - 1]
structure = structure + [0] * (self.max_elem_length + 2 - len(structure))
structure = np.array(structure)
data['structure'] = structure
elem_char_idx1 = self.dict_elem['<td>']
elem_char_idx2 = self.dict_elem['<td']
span_idx_list = self.get_span_idx_list()
td_idx_list = np.logical_or(structure == elem_char_idx1, structure == elem_char_idx2)
td_idx_list = np.where(td_idx_list)[0]
structure_mask = np.ones((self.max_elem_length + 2, 1), dtype=np.float32)
bbox_list = np.zeros((self.max_elem_length + 2, 4), dtype=np.float32)
bbox_list_mask = np.zeros((self.max_elem_length + 2, 1), dtype=np.float32)
img_height, img_width, img_ch = data['image'].shape
if len(span_idx_list) > 0:
span_weight = len(td_idx_list) * 1.0 / len(span_idx_list)
span_weight = min(max(span_weight, 1.0), self.span_weight)
for cno in range(len(cells)):
if 'bbox' in cells[cno]:
bbox = cells[cno]['bbox'].copy()
bbox[0] = bbox[0] * 1.0 / img_width
bbox[1] = bbox[1] * 1.0 / img_height
bbox[2] = bbox[2] * 1.0 / img_width
bbox[3] = bbox[3] * 1.0 / img_height
td_idx = td_idx_list[cno]
bbox_list[td_idx] = bbox
bbox_list_mask[td_idx] = 1.0
cand_span_idx = td_idx + 1
if cand_span_idx < (self.max_elem_length + 2):
if structure[cand_span_idx] in span_idx_list:
structure_mask[cand_span_idx] = span_weight
data['bbox_list'] = bbox_list
data['bbox_list_mask'] = bbox_list_mask
data['structure_mask'] = structure_mask
char_beg_idx = self.get_beg_end_flag_idx('beg', 'char')
char_end_idx = self.get_beg_end_flag_idx('end', 'char')
elem_beg_idx = self.get_beg_end_flag_idx('beg', 'elem')
elem_end_idx = self.get_beg_end_flag_idx('end', 'elem')
data['sp_tokens'] = np.array([char_beg_idx, char_end_idx, elem_beg_idx,
elem_end_idx, elem_char_idx1, elem_char_idx2, self.max_text_length,
self.max_elem_length, self.max_cell_num, elem_num])
return data
def encode(self, text, char_or_elem):
"""convert text-label into text-index.
"""
if char_or_elem == "char":
max_len = self.max_text_length
current_dict = self.dict_character
else:
max_len = self.max_elem_length
current_dict = self.dict_elem
if len(text) > max_len:
return None
if len(text) == 0:
if char_or_elem == "char":
return [self.dict_character['space']]
else:
return None
text_list = []
for char in text:
if char not in current_dict:
return None
text_list.append(current_dict[char])
if len(text_list) == 0:
if char_or_elem == "char":
return [self.dict_character['space']]
else:
return None
return text_list
def get_ignored_tokens(self, char_or_elem):
beg_idx = self.get_beg_end_flag_idx("beg", char_or_elem)
end_idx = self.get_beg_end_flag_idx("end", char_or_elem)
return [beg_idx, end_idx]
def get_beg_end_flag_idx(self, beg_or_end, char_or_elem):
if char_or_elem == "char":
if beg_or_end == "beg":
idx = np.array(self.dict_character[self.beg_str])
elif beg_or_end == "end":
idx = np.array(self.dict_character[self.end_str])
else:
assert False, "Unsupport type %s in get_beg_end_flag_idx of char" \
% beg_or_end
elif char_or_elem == "elem":
if beg_or_end == "beg":
idx = np.array(self.dict_elem[self.beg_str])
elif beg_or_end == "end":
idx = np.array(self.dict_elem[self.end_str])
else:
assert False, "Unsupport type %s in get_beg_end_flag_idx of elem" \
% beg_or_end
else:
assert False, "Unsupport type %s in char_or_elem" \
% char_or_elem
return idx
\ No newline at end of file
ppocr/data/imaug/operators.py
浏览文件 @ 253b8453
......@@ -163,7 +163,7 @@ class DetResizeForTest(object):
img, (ratio_h, ratio_w)
"""
limit_side_len = self.limit_side_len
h, w, _ = img.shape
h, w, c = img.shape
# limit the max side
if self.limit_type == 'max':
......@@ -174,7 +174,7 @@ class DetResizeForTest(object):
ratio = float(limit_side_len) / w
else:
ratio = 1.
else:
elif self.limit_type == 'min':
if min(h, w) < limit_side_len:
if h < w:
ratio = float(limit_side_len) / h
......@@ -182,6 +182,10 @@ class DetResizeForTest(object):
ratio = float(limit_side_len) / w
else:
ratio = 1.
elif self.limit_type == 'resize_long':
ratio = float(limit_side_len) / max(h,w)
else:
raise Exception('not support limit type, image ')
resize_h = int(h * ratio)
resize_w = int(w * ratio)
......
ppocr/data/pubtab_dataset.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
import os
import random
from paddle.io import Dataset
import json
from .imaug import transform, create_operators
class PubTabDataSet(Dataset):
def __init__(self, config, mode, logger, seed=None):
super(PubTabDataSet, self).__init__()
self.logger = logger
global_config = config['Global']
dataset_config = config[mode]['dataset']
loader_config = config[mode]['loader']
label_file_path = dataset_config.pop('label_file_path')
self.data_dir = dataset_config['data_dir']
self.do_shuffle = loader_config['shuffle']
self.do_hard_select = False
if 'hard_select' in loader_config:
self.do_hard_select = loader_config['hard_select']
self.hard_prob = loader_config['hard_prob']
if self.do_hard_select:
self.img_select_prob = self.load_hard_select_prob()
self.table_select_type = None
if 'table_select_type' in loader_config:
self.table_select_type = loader_config['table_select_type']
self.table_select_prob = loader_config['table_select_prob']
self.seed = seed
logger.info("Initialize indexs of datasets:%s" % label_file_path)
with open(label_file_path, "rb") as f:
self.data_lines = f.readlines()
self.data_idx_order_list = list(range(len(self.data_lines)))
if mode.lower() == "train":
self.shuffle_data_random()
self.ops = create_operators(dataset_config['transforms'], global_config)
def shuffle_data_random(self):
if self.do_shuffle:
random.seed(self.seed)
random.shuffle(self.data_lines)
return
def __getitem__(self, idx):
try:
data_line = self.data_lines[idx]
data_line = data_line.decode('utf-8').strip("\n")
info = json.loads(data_line)
file_name = info['filename']
select_flag = True
if self.do_hard_select:
prob = self.img_select_prob[file_name]
if prob < random.uniform(0, 1):
select_flag = False
if self.table_select_type:
structure = info['html']['structure']['tokens'].copy()
structure_str = ''.join(structure)
table_type = "simple"
if 'colspan' in structure_str or 'rowspan' in structure_str:
table_type = "complex"
if table_type == "complex":
if self.table_select_prob < random.uniform(0, 1):
select_flag = False
if select_flag:
cells = info['html']['cells'].copy()
structure = info['html']['structure'].copy()
img_path = os.path.join(self.data_dir, file_name)
data = {'img_path': img_path, 'cells': cells, 'structure':structure}
if not os.path.exists(img_path):
raise Exception("{} does not exist!".format(img_path))
with open(data['img_path'], 'rb') as f:
img = f.read()
data['image'] = img
outs = transform(data, self.ops)
else:
outs = None
except Exception as e:
self.logger.error(
"When parsing line {}, error happened with msg: {}".format(
data_line, e))
outs = None
if outs is None:
return self.__getitem__(np.random.randint(self.__len__()))
return outs
def __len__(self):
return len(self.data_idx_order_list)
ppocr/losses/__init__.py
浏览文件 @ 253b8453
......@@ -13,28 +13,39 @@
# limitations under the License.
import copy
import paddle
import paddle.nn as nn
# det loss
from .det_db_loss import DBLoss
from .det_east_loss import EASTLoss
from .det_sast_loss import SASTLoss
def build_loss(config):
# det loss
from .det_db_loss import DBLoss
from .det_east_loss import EASTLoss
from .det_sast_loss import SASTLoss
# rec loss
from .rec_ctc_loss import CTCLoss
from .rec_att_loss import AttentionLoss
from .rec_srn_loss import SRNLoss
# cls loss
from .cls_loss import ClsLoss
# e2e loss
from .e2e_pg_loss import PGLoss
# rec loss
from .rec_ctc_loss import CTCLoss
from .rec_att_loss import AttentionLoss
from .rec_srn_loss import SRNLoss
# basic loss function
from .basic_loss import DistanceLoss
# cls loss
from .cls_loss import ClsLoss
# combined loss function
from .combined_loss import CombinedLoss
# e2e loss
from .e2e_pg_loss import PGLoss
# table loss
from .table_att_loss import TableAttentionLoss
def build_loss(config):
support_dict = [
'DBLoss', 'EASTLoss', 'SASTLoss', 'CTCLoss', 'ClsLoss', 'AttentionLoss',
'SRNLoss', 'PGLoss']
'SRNLoss', 'PGLoss', 'CombinedLoss', 'TableAttentionLoss'
]
config = copy.deepcopy(config)
module_name = config.pop('name')
assert module_name in support_dict, Exception('loss only support {}'.format(
......
ppocr/losses/basic_loss.py 0 → 100644
浏览文件 @ 253b8453
#copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from paddle.nn import L1Loss
from paddle.nn import MSELoss as L2Loss
from paddle.nn import SmoothL1Loss
class CELoss(nn.Layer):
def __init__(self, epsilon=None):
super().__init__()
if epsilon is not None and (epsilon <= 0 or epsilon >= 1):
epsilon = None
self.epsilon = epsilon
def _labelsmoothing(self, target, class_num):
if target.shape[-1] != class_num:
one_hot_target = F.one_hot(target, class_num)
else:
one_hot_target = target
soft_target = F.label_smooth(one_hot_target, epsilon=self.epsilon)
soft_target = paddle.reshape(soft_target, shape=[-1, class_num])
return soft_target
def forward(self, x, label):
loss_dict = {}
if self.epsilon is not None:
class_num = x.shape[-1]
label = self._labelsmoothing(label, class_num)
x = -F.log_softmax(x, axis=-1)
loss = paddle.sum(x * label, axis=-1)
else:
if label.shape[-1] == x.shape[-1]:
label = F.softmax(label, axis=-1)
soft_label = True
else:
soft_label = False
loss = F.cross_entropy(x, label=label, soft_label=soft_label)
return loss
class DMLLoss(nn.Layer):
"""
DMLLoss
"""
def __init__(self, act=None):
super().__init__()
if act is not None:
assert act in ["softmax", "sigmoid"]
if act == "softmax":
self.act = nn.Softmax(axis=-1)
elif act == "sigmoid":
self.act = nn.Sigmoid()
else:
self.act = None
def forward(self, out1, out2):
if self.act is not None:
out1 = self.act(out1)
out2 = self.act(out2)
log_out1 = paddle.log(out1)
log_out2 = paddle.log(out2)
loss = (F.kl_div(
log_out1, out2, reduction='batchmean') + F.kl_div(
log_out2, out1, reduction='batchmean')) / 2.0
return loss
class DistanceLoss(nn.Layer):
"""
DistanceLoss:
mode: loss mode
"""
def __init__(self, mode="l2", **kargs):
super().__init__()
assert mode in ["l1", "l2", "smooth_l1"]
if mode == "l1":
self.loss_func = nn.L1Loss(**kargs)
elif mode == "l2":
self.loss_func = nn.MSELoss(**kargs)
elif mode == "smooth_l1":
self.loss_func = nn.SmoothL1Loss(**kargs)
def forward(self, x, y):
return self.loss_func(x, y)
ppocr/losses/cls_loss.py
浏览文件 @ 253b8453
......@@ -24,7 +24,7 @@ class ClsLoss(nn.Layer):
super(ClsLoss, self).__init__()
self.loss_func = nn.CrossEntropyLoss(reduction='mean')
def __call__(self, predicts, batch):
def forward(self, predicts, batch):
label = batch[1]
loss = self.loss_func(input=predicts, label=label)
return {'loss': loss}
ppocr/losses/combined_loss.py 0 → 100644
浏览文件 @ 253b8453
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import paddle
import paddle.nn as nn
from .distillation_loss import DistillationCTCLoss
from .distillation_loss import DistillationDMLLoss
from .distillation_loss import DistillationDistanceLoss
class CombinedLoss(nn.Layer):
"""
CombinedLoss:
a combionation of loss function
"""
def __init__(self, loss_config_list=None):
super().__init__()
self.loss_func = []
self.loss_weight = []
assert isinstance(loss_config_list, list), (
'operator config should be a list')
for config in loss_config_list:
assert isinstance(config,
dict) and len(config) == 1, "yaml format error"
name = list(config)[0]
param = config[name]
assert "weight" in param, "weight must be in param, but param just contains {}".format(
param.keys())
self.loss_weight.append(param.pop("weight"))
self.loss_func.append(eval(name)(**param))
def forward(self, input, batch, **kargs):
loss_dict = {}
for idx, loss_func in enumerate(self.loss_func):
loss = loss_func(input, batch, **kargs)
if isinstance(loss, paddle.Tensor):
loss = {"loss_{}_{}".format(str(loss), idx): loss}
weight = self.loss_weight[idx]
loss = {
"{}_{}".format(key, idx): loss[key] * weight
for key in loss
}
loss_dict.update(loss)
loss_dict["loss"] = paddle.add_n(list(loss_dict.values()))
return loss_dict
ppocr/losses/distillation_loss.py 0 → 100644
浏览文件 @ 253b8453
#copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
import paddle
import paddle.nn as nn
from .rec_ctc_loss import CTCLoss
from .basic_loss import DMLLoss
from .basic_loss import DistanceLoss
class DistillationDMLLoss(DMLLoss):
"""
"""
def __init__(self, model_name_pairs=[], act=None, key=None,
name="loss_dml"):
super().__init__(act=act)
assert isinstance(model_name_pairs, list)
self.key = key
self.model_name_pairs = model_name_pairs
self.name = name
def forward(self, predicts, batch):
loss_dict = dict()
for idx, pair in enumerate(self.model_name_pairs):
out1 = predicts[pair[0]]
out2 = predicts[pair[1]]
if self.key is not None:
out1 = out1[self.key]
out2 = out2[self.key]
loss = super().forward(out1, out2)
if isinstance(loss, dict):
for key in loss:
loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1],
idx)] = loss[key]
else:
loss_dict["{}_{}".format(self.name, idx)] = loss
return loss_dict
class DistillationCTCLoss(CTCLoss):
def __init__(self, model_name_list=[], key=None, name="loss_ctc"):
super().__init__()
self.model_name_list = model_name_list
self.key = key
self.name = name
def forward(self, predicts, batch):
loss_dict = dict()
for idx, model_name in enumerate(self.model_name_list):
out = predicts[model_name]
if self.key is not None:
out = out[self.key]
loss = super().forward(out, batch)
if isinstance(loss, dict):
for key in loss:
loss_dict["{}_{}_{}".format(self.name, model_name,
idx)] = loss[key]
else:
loss_dict["{}_{}".format(self.name, model_name)] = loss
return loss_dict
class DistillationDistanceLoss(DistanceLoss):
"""
"""
def __init__(self,
mode="l2",
model_name_pairs=[],
key=None,
name="loss_distance",
**kargs):
super().__init__(mode=mode, **kargs)
assert isinstance(model_name_pairs, list)
self.key = key
self.model_name_pairs = model_name_pairs
self.name = name + "_l2"
def forward(self, predicts, batch):
loss_dict = dict()
for idx, pair in enumerate(self.model_name_pairs):
out1 = predicts[pair[0]]
out2 = predicts[pair[1]]
if self.key is not None:
out1 = out1[self.key]
out2 = out2[self.key]
loss = super().forward(out1, out2)
if isinstance(loss, dict):
for key in loss:
loss_dict["{}_{}_{}".format(self.name, key, idx)] = loss[
key]
else:
loss_dict["{}_{}_{}_{}".format(self.name, pair[0], pair[1],
idx)] = loss
return loss_dict
ppocr/losses/rec_ctc_loss.py
浏览文件 @ 253b8453
......@@ -25,7 +25,7 @@ class CTCLoss(nn.Layer):
super(CTCLoss, self).__init__()
self.loss_func = nn.CTCLoss(blank=0, reduction='none')
def __call__(self, predicts, batch):
def forward(self, predicts, batch):
predicts = predicts.transpose((1, 0, 2))
N, B, _ = predicts.shape
preds_lengths = paddle.to_tensor([N] * B, dtype='int64')
......
ppocr/losses/table_att_loss.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
from paddle import nn
from paddle.nn import functional as F
from paddle import fluid
class TableAttentionLoss(nn.Layer):
def __init__(self, structure_weight, loc_weight, use_giou=False, giou_weight=1.0, **kwargs):
super(TableAttentionLoss, self).__init__()
self.loss_func = nn.CrossEntropyLoss(weight=None, reduction='none')
self.structure_weight = structure_weight
self.loc_weight = loc_weight
self.use_giou = use_giou
self.giou_weight = giou_weight
def giou_loss(self, preds, bbox, eps=1e-7, reduction='mean'):
'''
:param preds:[[x1,y1,x2,y2], [x1,y1,x2,y2],,,]
:param bbox:[[x1,y1,x2,y2], [x1,y1,x2,y2],,,]
:return: loss
'''
ix1 = fluid.layers.elementwise_max(preds[:, 0], bbox[:, 0])
iy1 = fluid.layers.elementwise_max(preds[:, 1], bbox[:, 1])
ix2 = fluid.layers.elementwise_min(preds[:, 2], bbox[:, 2])
iy2 = fluid.layers.elementwise_min(preds[:, 3], bbox[:, 3])
iw = fluid.layers.clip(ix2 - ix1 + 1e-3, 0., 1e10)
ih = fluid.layers.clip(iy2 - iy1 + 1e-3, 0., 1e10)
# overlap
inters = iw * ih
# union
uni = (preds[:, 2] - preds[:, 0] + 1e-3) * (preds[:, 3] - preds[:, 1] + 1e-3
) + (bbox[:, 2] - bbox[:, 0] + 1e-3) * (
bbox[:, 3] - bbox[:, 1] + 1e-3) - inters + eps
# ious
ious = inters / uni
ex1 = fluid.layers.elementwise_min(preds[:, 0], bbox[:, 0])
ey1 = fluid.layers.elementwise_min(preds[:, 1], bbox[:, 1])
ex2 = fluid.layers.elementwise_max(preds[:, 2], bbox[:, 2])
ey2 = fluid.layers.elementwise_max(preds[:, 3], bbox[:, 3])
ew = fluid.layers.clip(ex2 - ex1 + 1e-3, 0., 1e10)
eh = fluid.layers.clip(ey2 - ey1 + 1e-3, 0., 1e10)
# enclose erea
enclose = ew * eh + eps
giou = ious - (enclose - uni) / enclose
loss = 1 - giou
if reduction == 'mean':
loss = paddle.mean(loss)
elif reduction == 'sum':
loss = paddle.sum(loss)
else:
raise NotImplementedError
return loss
def forward(self, predicts, batch):
structure_probs = predicts['structure_probs']
structure_targets = batch[1].astype("int64")
structure_targets = structure_targets[:, 1:]
if len(batch) == 6:
structure_mask = batch[5].astype("int64")
structure_mask = structure_mask[:, 1:]
structure_mask = paddle.reshape(structure_mask, [-1])
structure_probs = paddle.reshape(structure_probs, [-1, structure_probs.shape[-1]])
structure_targets = paddle.reshape(structure_targets, [-1])
structure_loss = self.loss_func(structure_probs, structure_targets)
if len(batch) == 6:
structure_loss = structure_loss * structure_mask
# structure_loss = paddle.sum(structure_loss) * self.structure_weight
structure_loss = paddle.mean(structure_loss) * self.structure_weight
loc_preds = predicts['loc_preds']
loc_targets = batch[2].astype("float32")
loc_targets_mask = batch[4].astype("float32")
loc_targets = loc_targets[:, 1:, :]
loc_targets_mask = loc_targets_mask[:, 1:, :]
loc_loss = F.mse_loss(loc_preds * loc_targets_mask, loc_targets) * self.loc_weight
if self.use_giou:
loc_loss_giou = self.giou_loss(loc_preds * loc_targets_mask, loc_targets) * self.giou_weight
total_loss = structure_loss + loc_loss + loc_loss_giou
return {'loss':total_loss, "structure_loss":structure_loss, "loc_loss":loc_loss, "loc_loss_giou":loc_loss_giou}
else:
total_loss = structure_loss + loc_loss
return {'loss':total_loss, "structure_loss":structure_loss, "loc_loss":loc_loss}
\ No newline at end of file
ppocr/metrics/__init__.py
浏览文件 @ 253b8453
......@@ -19,20 +19,23 @@ from __future__ import unicode_literals
import copy
__all__ = ['build_metric']
__all__ = ["build_metric"]
from .det_metric import DetMetric
from .rec_metric import RecMetric
from .cls_metric import ClsMetric
from .e2e_metric import E2EMetric
from .distillation_metric import DistillationMetric
from .table_metric import TableMetric
def build_metric(config):
from .det_metric import DetMetric
from .rec_metric import RecMetric
from .cls_metric import ClsMetric
from .e2e_metric import E2EMetric
support_dict = ['DetMetric', 'RecMetric', 'ClsMetric', 'E2EMetric']
support_dict = [
"DetMetric", "RecMetric", "ClsMetric", "E2EMetric", "DistillationMetric", "TableMetric"
]
config = copy.deepcopy(config)
module_name = config.pop('name')
module_name = config.pop("name")
assert module_name in support_dict, Exception(
'metric only support {}'.format(support_dict))
"metric only support {}".format(support_dict))
module_class = eval(module_name)(**config)
return module_class
ppocr/metrics/distillation_metric.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import importlib
import copy
from .rec_metric import RecMetric
from .det_metric import DetMetric
from .e2e_metric import E2EMetric
from .cls_metric import ClsMetric
class DistillationMetric(object):
def __init__(self,
key=None,
base_metric_name="RecMetric",
main_indicator='acc',
**kwargs):
self.main_indicator = main_indicator
self.key = key
self.main_indicator = main_indicator
self.base_metric_name = base_metric_name
self.kwargs = kwargs
self.metrics = None
def _init_metrcis(self, preds):
self.metrics = dict()
mod = importlib.import_module(__name__)
for key in preds:
self.metrics[key] = getattr(mod, self.base_metric_name)(
main_indicator=self.main_indicator, **self.kwargs)
self.metrics[key].reset()
def __call__(self, preds, *args, **kwargs):
assert isinstance(preds, dict)
if self.metrics is None:
self._init_metrcis(preds)
output = dict()
for key in preds:
metric = self.metrics[key].__call__(preds[key], *args, **kwargs)
for sub_key in metric:
output["{}_{}".format(key, sub_key)] = metric[sub_key]
return output
def get_metric(self):
"""
return metrics {
'acc': 0,
'norm_edit_dis': 0,
}
"""
output = dict()
for key in self.metrics:
metric = self.metrics[key].get_metric()
# main indicator
if key == self.key:
output.update(metric)
else:
for sub_key in metric:
output["{}_{}".format(key, sub_key)] = metric[sub_key]
return output
def reset(self):
for key in self.metrics:
self.metrics[key].reset()
ppocr/metrics/table_metric.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
class TableMetric(object):
def __init__(self, main_indicator='acc', **kwargs):
self.main_indicator = main_indicator
self.reset()
def __call__(self, pred, batch, *args, **kwargs):
structure_probs = pred['structure_probs'].numpy()
structure_labels = batch[1]
correct_num = 0
all_num = 0
structure_probs = np.argmax(structure_probs, axis=2)
structure_labels = structure_labels[:, 1:]
batch_size = structure_probs.shape[0]
for bno in range(batch_size):
all_num += 1
if (structure_probs[bno] == structure_labels[bno]).all():
correct_num += 1
self.correct_num += correct_num
self.all_num += all_num
return {
'acc': correct_num * 1.0 / all_num,
}
def get_metric(self):
"""
return metrics {
'acc': 0,
}
"""
acc = 1.0 * self.correct_num / self.all_num
self.reset()
return {'acc': acc}
def reset(self):
self.correct_num = 0
self.all_num = 0
ppocr/modeling/architectures/__init__.py
浏览文件 @ 253b8453
......@@ -13,12 +13,20 @@
# limitations under the License.
import copy
import importlib
from .base_model import BaseModel
from .distillation_model import DistillationModel
__all__ = ['build_model']
def build_model(config):
from .base_model import BaseModel
def build_model(config):
config = copy.deepcopy(config)
module_class = BaseModel(config)
return module_class
\ No newline at end of file
if not "name" in config:
arch = BaseModel(config)
else:
name = config.pop("name")
mod = importlib.import_module(__name__)
arch = getattr(mod, name)(config)
return arch
ppocr/modeling/architectures/base_model.py
浏览文件 @ 253b8453
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
......@@ -32,7 +32,6 @@ class BaseModel(nn.Layer):
config (dict): the super parameters for module.
"""
super(BaseModel, self).__init__()
in_channels = config.get('in_channels', 3)
model_type = config['model_type']
# build transfrom,
......@@ -68,14 +67,20 @@ class BaseModel(nn.Layer):
config["Head"]['in_channels'] = in_channels
self.head = build_head(config["Head"])
self.return_all_feats = config.get("return_all_feats", False)
def forward(self, x, data=None):
y = dict()
if self.use_transform:
x = self.transform(x)
x = self.backbone(x)
y["backbone_out"] = x
if self.use_neck:
x = self.neck(x)
if data is None:
x = self.head(x)
y["neck_out"] = x
x = self.head(x, targets=data)
y["head_out"] = x
if self.return_all_feats:
return y
else:
x = self.head(x, data)
return x
ppocr/modeling/architectures/distillation_model.py 0 → 100644
浏览文件 @ 253b8453
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from paddle import nn
from ppocr.modeling.transforms import build_transform
from ppocr.modeling.backbones import build_backbone
from ppocr.modeling.necks import build_neck
from ppocr.modeling.heads import build_head
from .base_model import BaseModel
from ppocr.utils.save_load import init_model
__all__ = ['DistillationModel']
class DistillationModel(nn.Layer):
def __init__(self, config):
"""
the module for OCR distillation.
args:
config (dict): the super parameters for module.
"""
super().__init__()
self.model_list = []
self.model_name_list = []
for key in config["Models"]:
model_config = config["Models"][key]
freeze_params = False
pretrained = None
if "freeze_params" in model_config:
freeze_params = model_config.pop("freeze_params")
if "pretrained" in model_config:
pretrained = model_config.pop("pretrained")
model = BaseModel(model_config)
if pretrained is not None:
init_model(model, path=pretrained)
if freeze_params:
for param in model.parameters():
param.trainable = False
self.model_list.append(self.add_sublayer(key, model))
self.model_name_list.append(key)
def forward(self, x):
result_dict = dict()
for idx, model_name in enumerate(self.model_name_list):
result_dict[model_name] = self.model_list[idx](x)
return result_dict
ppocr/modeling/backbones/__init__.py
浏览文件 @ 253b8453
......@@ -29,6 +29,10 @@ def build_backbone(config, model_type):
elif model_type == 'e2e':
from .e2e_resnet_vd_pg import ResNet
support_dict = ['ResNet']
elif model_type == "table":
from .table_resnet_vd import ResNet
from .table_mobilenet_v3 import MobileNetV3
support_dict = ['ResNet', 'MobileNetV3']
else:
raise NotImplementedError
......
ppocr/modeling/backbones/det_mobilenet_v3.py
浏览文件 @ 253b8453
......@@ -102,8 +102,7 @@ class MobileNetV3(nn.Layer):
padding=1,
groups=1,
if_act=True,
act='hardswish',
name='conv1')
act='hardswish')
self.stages = []
self.out_channels = []
......@@ -125,8 +124,7 @@ class MobileNetV3(nn.Layer):
kernel_size=k,
stride=s,
use_se=se,
act=nl,
name="conv" + str(i + 2)))
act=nl))
inplanes = make_divisible(scale * c)
i += 1
block_list.append(
......@@ -138,8 +136,7 @@ class MobileNetV3(nn.Layer):
padding=0,
groups=1,
if_act=True,
act='hardswish',
name='conv_last'))
act='hardswish'))
self.stages.append(nn.Sequential(*block_list))
self.out_channels.append(make_divisible(scale * cls_ch_squeeze))
for i, stage in enumerate(self.stages):
......@@ -163,8 +160,7 @@ class ConvBNLayer(nn.Layer):
padding,
groups=1,
if_act=True,
act=None,
name=None):
act=None):
super(ConvBNLayer, self).__init__()
self.if_act = if_act
self.act = act
......@@ -175,16 +171,9 @@ class ConvBNLayer(nn.Layer):
stride=stride,
padding=padding,
groups=groups,
weight_attr=ParamAttr(name=name + '_weights'),
bias_attr=False)
self.bn = nn.BatchNorm(
num_channels=out_channels,
act=None,
param_attr=ParamAttr(name=name + "_bn_scale"),
bias_attr=ParamAttr(name=name + "_bn_offset"),
moving_mean_name=name + "_bn_mean",
moving_variance_name=name + "_bn_variance")
self.bn = nn.BatchNorm(num_channels=out_channels, act=None)
def forward(self, x):
x = self.conv(x)
......@@ -209,8 +198,7 @@ class ResidualUnit(nn.Layer):
kernel_size,
stride,
use_se,
act=None,
name=''):
act=None):
super(ResidualUnit, self).__init__()
self.if_shortcut = stride == 1 and in_channels == out_channels
self.if_se = use_se
......@@ -222,8 +210,7 @@ class ResidualUnit(nn.Layer):
stride=1,
padding=0,
if_act=True,
act=act,
name=name + "_expand")
act=act)
self.bottleneck_conv = ConvBNLayer(
in_channels=mid_channels,
out_channels=mid_channels,
......@@ -232,10 +219,9 @@ class ResidualUnit(nn.Layer):
padding=int((kernel_size - 1) // 2),
groups=mid_channels,
if_act=True,
act=act,
name=name + "_depthwise")
act=act)
if self.if_se:
self.mid_se = SEModule(mid_channels, name=name + "_se")
self.mid_se = SEModule(mid_channels)
self.linear_conv = ConvBNLayer(
in_channels=mid_channels,
out_channels=out_channels,
......@@ -243,8 +229,7 @@ class ResidualUnit(nn.Layer):
stride=1,
padding=0,
if_act=False,
act=None,
name=name + "_linear")
act=None)
def forward(self, inputs):
x = self.expand_conv(inputs)
......@@ -258,7 +243,7 @@ class ResidualUnit(nn.Layer):
class SEModule(nn.Layer):
def __init__(self, in_channels, reduction=4, name=""):
def __init__(self, in_channels, reduction=4):
super(SEModule, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2D(1)
self.conv1 = nn.Conv2D(
......@@ -266,17 +251,13 @@ class SEModule(nn.Layer):
out_channels=in_channels // reduction,
kernel_size=1,
stride=1,
padding=0,
weight_attr=ParamAttr(name=name + "_1_weights"),
bias_attr=ParamAttr(name=name + "_1_offset"))
padding=0)
self.conv2 = nn.Conv2D(
in_channels=in_channels // reduction,
out_channels=in_channels,
kernel_size=1,
stride=1,
padding=0,
weight_attr=ParamAttr(name + "_2_weights"),
bias_attr=ParamAttr(name=name + "_2_offset"))
padding=0)
def forward(self, inputs):
outputs = self.avg_pool(inputs)
......
ppocr/modeling/backbones/rec_mobilenet_v3.py
浏览文件 @ 253b8453
......@@ -96,8 +96,7 @@ class MobileNetV3(nn.Layer):
padding=1,
groups=1,
if_act=True,
act='hardswish',
name='conv1')
act='hardswish')
i = 0
block_list = []
inplanes = make_divisible(inplanes * scale)
......@@ -110,8 +109,7 @@ class MobileNetV3(nn.Layer):
kernel_size=k,
stride=s,
use_se=se,
act=nl,
name='conv' + str(i + 2)))
act=nl))
inplanes = make_divisible(scale * c)
i += 1
self.blocks = nn.Sequential(*block_list)
......@@ -124,8 +122,7 @@ class MobileNetV3(nn.Layer):
padding=0,
groups=1,
if_act=True,
act='hardswish',
name='conv_last')
act='hardswish')
self.pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
self.out_channels = make_divisible(scale * cls_ch_squeeze)
......
ppocr/modeling/backbones/table_mobilenet_v3.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
from paddle import nn
import paddle.nn.functional as F
from paddle import ParamAttr
__all__ = ['MobileNetV3']
def make_divisible(v, divisor=8, min_value=None):
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
if new_v < 0.9 * v:
new_v += divisor
return new_v
class MobileNetV3(nn.Layer):
def __init__(self,
in_channels=3,
model_name='large',
scale=0.5,
disable_se=False,
**kwargs):
"""
the MobilenetV3 backbone network for detection module.
Args:
params(dict): the super parameters for build network
"""
super(MobileNetV3, self).__init__()
self.disable_se = disable_se
if model_name == "large":
cfg = [
# k, exp, c, se, nl, s,
[3, 16, 16, False, 'relu', 1],
[3, 64, 24, False, 'relu', 2],
[3, 72, 24, False, 'relu', 1],
[5, 72, 40, True, 'relu', 2],
[5, 120, 40, True, 'relu', 1],
[5, 120, 40, True, 'relu', 1],
[3, 240, 80, False, 'hardswish', 2],
[3, 200, 80, False, 'hardswish', 1],
[3, 184, 80, False, 'hardswish', 1],
[3, 184, 80, False, 'hardswish', 1],
[3, 480, 112, True, 'hardswish', 1],
[3, 672, 112, True, 'hardswish', 1],
[5, 672, 160, True, 'hardswish', 2],
[5, 960, 160, True, 'hardswish', 1],
[5, 960, 160, True, 'hardswish', 1],
]
cls_ch_squeeze = 960
elif model_name == "small":
cfg = [
# k, exp, c, se, nl, s,
[3, 16, 16, True, 'relu', 2],
[3, 72, 24, False, 'relu', 2],
[3, 88, 24, False, 'relu', 1],
[5, 96, 40, True, 'hardswish', 2],
[5, 240, 40, True, 'hardswish', 1],
[5, 240, 40, True, 'hardswish', 1],
[5, 120, 48, True, 'hardswish', 1],
[5, 144, 48, True, 'hardswish', 1],
[5, 288, 96, True, 'hardswish', 2],
[5, 576, 96, True, 'hardswish', 1],
[5, 576, 96, True, 'hardswish', 1],
]
cls_ch_squeeze = 576
else:
raise NotImplementedError("mode[" + model_name +
"_model] is not implemented!")
supported_scale = [0.35, 0.5, 0.75, 1.0, 1.25]
assert scale in supported_scale, \
"supported scale are {} but input scale is {}".format(supported_scale, scale)
inplanes = 16
# conv1
self.conv = ConvBNLayer(
in_channels=in_channels,
out_channels=make_divisible(inplanes * scale),
kernel_size=3,
stride=2,
padding=1,
groups=1,
if_act=True,
act='hardswish',
name='conv1')
self.stages = []
self.out_channels = []
block_list = []
i = 0
inplanes = make_divisible(inplanes * scale)
for (k, exp, c, se, nl, s) in cfg:
se = se and not self.disable_se
start_idx = 2 if model_name == 'large' else 0
if s == 2 and i > start_idx:
self.out_channels.append(inplanes)
self.stages.append(nn.Sequential(*block_list))
block_list = []
block_list.append(
ResidualUnit(
in_channels=inplanes,
mid_channels=make_divisible(scale * exp),
out_channels=make_divisible(scale * c),
kernel_size=k,
stride=s,
use_se=se,
act=nl,
name="conv" + str(i + 2)))
inplanes = make_divisible(scale * c)
i += 1
block_list.append(
ConvBNLayer(
in_channels=inplanes,
out_channels=make_divisible(scale * cls_ch_squeeze),
kernel_size=1,
stride=1,
padding=0,
groups=1,
if_act=True,
act='hardswish',
name='conv_last'))
self.stages.append(nn.Sequential(*block_list))
self.out_channels.append(make_divisible(scale * cls_ch_squeeze))
for i, stage in enumerate(self.stages):
self.add_sublayer(sublayer=stage, name="stage{}".format(i))
def forward(self, x):
x = self.conv(x)
out_list = []
for stage in self.stages:
x = stage(x)
out_list.append(x)
return out_list
class ConvBNLayer(nn.Layer):
def __init__(self,
in_channels,
out_channels,
kernel_size,
stride,
padding,
groups=1,
if_act=True,
act=None,
name=None):
super(ConvBNLayer, self).__init__()
self.if_act = if_act
self.act = act
self.conv = nn.Conv2D(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
stride=stride,
padding=padding,
groups=groups,
weight_attr=ParamAttr(name=name + '_weights'),
bias_attr=False)
self.bn = nn.BatchNorm(
num_channels=out_channels,
act=None,
param_attr=ParamAttr(name=name + "_bn_scale"),
bias_attr=ParamAttr(name=name + "_bn_offset"),
moving_mean_name=name + "_bn_mean",
moving_variance_name=name + "_bn_variance")
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
if self.if_act:
if self.act == "relu":
x = F.relu(x)
elif self.act == "hardswish":
x = F.hardswish(x)
else:
print("The activation function({}) is selected incorrectly.".
format(self.act))
exit()
return x
class ResidualUnit(nn.Layer):
def __init__(self,
in_channels,
mid_channels,
out_channels,
kernel_size,
stride,
use_se,
act=None,
name=''):
super(ResidualUnit, self).__init__()
self.if_shortcut = stride == 1 and in_channels == out_channels
self.if_se = use_se
self.expand_conv = ConvBNLayer(
in_channels=in_channels,
out_channels=mid_channels,
kernel_size=1,
stride=1,
padding=0,
if_act=True,
act=act,
name=name + "_expand")
self.bottleneck_conv = ConvBNLayer(
in_channels=mid_channels,
out_channels=mid_channels,
kernel_size=kernel_size,
stride=stride,
padding=int((kernel_size - 1) // 2),
groups=mid_channels,
if_act=True,
act=act,
name=name + "_depthwise")
if self.if_se:
self.mid_se = SEModule(mid_channels, name=name + "_se")
self.linear_conv = ConvBNLayer(
in_channels=mid_channels,
out_channels=out_channels,
kernel_size=1,
stride=1,
padding=0,
if_act=False,
act=None,
name=name + "_linear")
def forward(self, inputs):
x = self.expand_conv(inputs)
x = self.bottleneck_conv(x)
if self.if_se:
x = self.mid_se(x)
x = self.linear_conv(x)
if self.if_shortcut:
x = paddle.add(inputs, x)
return x
class SEModule(nn.Layer):
def __init__(self, in_channels, reduction=4, name=""):
super(SEModule, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2D(1)
self.conv1 = nn.Conv2D(
in_channels=in_channels,
out_channels=in_channels // reduction,
kernel_size=1,
stride=1,
padding=0,
weight_attr=ParamAttr(name=name + "_1_weights"),
bias_attr=ParamAttr(name=name + "_1_offset"))
self.conv2 = nn.Conv2D(
in_channels=in_channels // reduction,
out_channels=in_channels,
kernel_size=1,
stride=1,
padding=0,
weight_attr=ParamAttr(name + "_2_weights"),
bias_attr=ParamAttr(name=name + "_2_offset"))
def forward(self, inputs):
outputs = self.avg_pool(inputs)
outputs = self.conv1(outputs)
outputs = F.relu(outputs)
outputs = self.conv2(outputs)
outputs = F.hardsigmoid(outputs, slope=0.2, offset=0.5)
return inputs * outputs
\ No newline at end of file
ppocr/modeling/backbones/table_resnet_vd.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
from paddle import ParamAttr
import paddle.nn as nn
import paddle.nn.functional as F
__all__ = ["ResNet"]
class ConvBNLayer(nn.Layer):
def __init__(
self,
in_channels,
out_channels,
kernel_size,
stride=1,
groups=1,
is_vd_mode=False,
act=None,
name=None, ):
super(ConvBNLayer, self).__init__()
self.is_vd_mode = is_vd_mode
self._pool2d_avg = nn.AvgPool2D(
kernel_size=2, stride=2, padding=0, ceil_mode=True)
self._conv = nn.Conv2D(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
stride=stride,
padding=(kernel_size - 1) // 2,
groups=groups,
weight_attr=ParamAttr(name=name + "_weights"),
bias_attr=False)
if name == "conv1":
bn_name = "bn_" + name
else:
bn_name = "bn" + name[3:]
self._batch_norm = nn.BatchNorm(
out_channels,
act=act,
param_attr=ParamAttr(name=bn_name + '_scale'),
bias_attr=ParamAttr(bn_name + '_offset'),
moving_mean_name=bn_name + '_mean',
moving_variance_name=bn_name + '_variance')
def forward(self, inputs):
if self.is_vd_mode:
inputs = self._pool2d_avg(inputs)
y = self._conv(inputs)
y = self._batch_norm(y)
return y
class BottleneckBlock(nn.Layer):
def __init__(self,
in_channels,
out_channels,
stride,
shortcut=True,
if_first=False,
name=None):
super(BottleneckBlock, self).__init__()
self.conv0 = ConvBNLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
act='relu',
name=name + "_branch2a")
self.conv1 = ConvBNLayer(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=3,
stride=stride,
act='relu',
name=name + "_branch2b")
self.conv2 = ConvBNLayer(
in_channels=out_channels,
out_channels=out_channels * 4,
kernel_size=1,
act=None,
name=name + "_branch2c")
if not shortcut:
self.short = ConvBNLayer(
in_channels=in_channels,
out_channels=out_channels * 4,
kernel_size=1,
stride=1,
is_vd_mode=False if if_first else True,
name=name + "_branch1")
self.shortcut = shortcut
def forward(self, inputs):
y = self.conv0(inputs)
conv1 = self.conv1(y)
conv2 = self.conv2(conv1)
if self.shortcut:
short = inputs
else:
short = self.short(inputs)
y = paddle.add(x=short, y=conv2)
y = F.relu(y)
return y
class BasicBlock(nn.Layer):
def __init__(self,
in_channels,
out_channels,
stride,
shortcut=True,
if_first=False,
name=None):
super(BasicBlock, self).__init__()
self.stride = stride
self.conv0 = ConvBNLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=3,
stride=stride,
act='relu',
name=name + "_branch2a")
self.conv1 = ConvBNLayer(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=3,
act=None,
name=name + "_branch2b")
if not shortcut:
self.short = ConvBNLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
stride=1,
is_vd_mode=False if if_first else True,
name=name + "_branch1")
self.shortcut = shortcut
def forward(self, inputs):
y = self.conv0(inputs)
conv1 = self.conv1(y)
if self.shortcut:
short = inputs
else:
short = self.short(inputs)
y = paddle.add(x=short, y=conv1)
y = F.relu(y)
return y
class ResNet(nn.Layer):
def __init__(self, in_channels=3, layers=50, **kwargs):
super(ResNet, self).__init__()
self.layers = layers
supported_layers = [18, 34, 50, 101, 152, 200]
assert layers in supported_layers, \
"supported layers are {} but input layer is {}".format(
supported_layers, layers)
if layers == 18:
depth = [2, 2, 2, 2]
elif layers == 34 or layers == 50:
depth = [3, 4, 6, 3]
elif layers == 101:
depth = [3, 4, 23, 3]
elif layers == 152:
depth = [3, 8, 36, 3]
elif layers == 200:
depth = [3, 12, 48, 3]
num_channels = [64, 256, 512,
1024] if layers >= 50 else [64, 64, 128, 256]
num_filters = [64, 128, 256, 512]
self.conv1_1 = ConvBNLayer(
in_channels=in_channels,
out_channels=32,
kernel_size=3,
stride=2,
act='relu',
name="conv1_1")
self.conv1_2 = ConvBNLayer(
in_channels=32,
out_channels=32,
kernel_size=3,
stride=1,
act='relu',
name="conv1_2")
self.conv1_3 = ConvBNLayer(
in_channels=32,
out_channels=64,
kernel_size=3,
stride=1,
act='relu',
name="conv1_3")
self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
self.stages = []
self.out_channels = []
if layers >= 50:
for block in range(len(depth)):
block_list = []
shortcut = False
for i in range(depth[block]):
if layers in [101, 152] and block == 2:
if i == 0:
conv_name = "res" + str(block + 2) + "a"
else:
conv_name = "res" + str(block + 2) + "b" + str(i)
else:
conv_name = "res" + str(block + 2) + chr(97 + i)
bottleneck_block = self.add_sublayer(
'bb_%d_%d' % (block, i),
BottleneckBlock(
in_channels=num_channels[block]
if i == 0 else num_filters[block] * 4,
out_channels=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
shortcut=shortcut,
if_first=block == i == 0,
name=conv_name))
shortcut = True
block_list.append(bottleneck_block)
self.out_channels.append(num_filters[block] * 4)
self.stages.append(nn.Sequential(*block_list))
else:
for block in range(len(depth)):
block_list = []
shortcut = False
for i in range(depth[block]):
conv_name = "res" + str(block + 2) + chr(97 + i)
basic_block = self.add_sublayer(
'bb_%d_%d' % (block, i),
BasicBlock(
in_channels=num_channels[block]
if i == 0 else num_filters[block],
out_channels=num_filters[block],
stride=2 if i == 0 and block != 0 else 1,
shortcut=shortcut,
if_first=block == i == 0,
name=conv_name))
shortcut = True
block_list.append(basic_block)
self.out_channels.append(num_filters[block])
self.stages.append(nn.Sequential(*block_list))
def forward(self, inputs):
y = self.conv1_1(inputs)
y = self.conv1_2(y)
y = self.conv1_3(y)
y = self.pool2d_max(y)
out = []
for block in self.stages:
y = block(y)
out.append(y)
return out
ppocr/modeling/heads/__init__.py
浏览文件 @ 253b8453
......@@ -31,8 +31,10 @@ def build_head(config):
from .cls_head import ClsHead
support_dict = [
'DBHead', 'EASTHead', 'SASTHead', 'CTCHead', 'ClsHead', 'AttentionHead',
'SRNHead', 'PGHead']
'SRNHead', 'PGHead', 'TableAttentionHead']
#table head
from .table_att_head import TableAttentionHead
module_name = config.pop('name')
assert module_name in support_dict, Exception('head only support {}'.format(
......
ppocr/modeling/heads/cls_head.py
浏览文件 @ 253b8453
......@@ -43,7 +43,7 @@ class ClsHead(nn.Layer):
initializer=nn.initializer.Uniform(-stdv, stdv)),
bias_attr=ParamAttr(name="fc_0.b_0"), )
def forward(self, x):
def forward(self, x, targets=None):
x = self.pool(x)
x = paddle.reshape(x, shape=[x.shape[0], x.shape[1]])
x = self.fc(x)
......
ppocr/modeling/heads/det_db_head.py
浏览文件 @ 253b8453
......@@ -23,10 +23,10 @@ import paddle.nn.functional as F
from paddle import ParamAttr
def get_bias_attr(k, name):
def get_bias_attr(k):
stdv = 1.0 / math.sqrt(k * 1.0)
initializer = paddle.nn.initializer.Uniform(-stdv, stdv)
bias_attr = ParamAttr(initializer=initializer, name=name + "_b_attr")
bias_attr = ParamAttr(initializer=initializer)
return bias_attr
......@@ -38,18 +38,14 @@ class Head(nn.Layer):
out_channels=in_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(name=name_list[0] + '.w_0'),
weight_attr=ParamAttr(),
bias_attr=False)
self.conv_bn1 = nn.BatchNorm(
num_channels=in_channels // 4,
param_attr=ParamAttr(
name=name_list[1] + '.w_0',
initializer=paddle.nn.initializer.Constant(value=1.0)),
bias_attr=ParamAttr(
name=name_list[1] + '.b_0',
initializer=paddle.nn.initializer.Constant(value=1e-4)),
moving_mean_name=name_list[1] + '.w_1',
moving_variance_name=name_list[1] + '.w_2',
act='relu')
self.conv2 = nn.Conv2DTranspose(
in_channels=in_channels // 4,
......@@ -57,19 +53,14 @@ class Head(nn.Layer):
kernel_size=2,
stride=2,
weight_attr=ParamAttr(
name=name_list[2] + '.w_0',
initializer=paddle.nn.initializer.KaimingUniform()),
bias_attr=get_bias_attr(in_channels // 4, name_list[-1] + "conv2"))
bias_attr=get_bias_attr(in_channels // 4))
self.conv_bn2 = nn.BatchNorm(
num_channels=in_channels // 4,
param_attr=ParamAttr(
name=name_list[3] + '.w_0',
initializer=paddle.nn.initializer.Constant(value=1.0)),
bias_attr=ParamAttr(
name=name_list[3] + '.b_0',
initializer=paddle.nn.initializer.Constant(value=1e-4)),
moving_mean_name=name_list[3] + '.w_1',
moving_variance_name=name_list[3] + '.w_2',
act="relu")
self.conv3 = nn.Conv2DTranspose(
in_channels=in_channels // 4,
......@@ -77,10 +68,8 @@ class Head(nn.Layer):
kernel_size=2,
stride=2,
weight_attr=ParamAttr(
name=name_list[4] + '.w_0',
initializer=paddle.nn.initializer.KaimingUniform()),
bias_attr=get_bias_attr(in_channels // 4, name_list[-1] + "conv3"),
)
bias_attr=get_bias_attr(in_channels // 4), )
def forward(self, x):
x = self.conv1(x)
......@@ -117,7 +106,7 @@ class DBHead(nn.Layer):
def step_function(self, x, y):
return paddle.reciprocal(1 + paddle.exp(-self.k * (x - y)))
def forward(self, x):
def forward(self, x, targets=None):
shrink_maps = self.binarize(x)
if not self.training:
return {'maps': shrink_maps}
......
ppocr/modeling/heads/det_east_head.py
浏览文件 @ 253b8453
......@@ -109,7 +109,7 @@ class EASTHead(nn.Layer):
act=None,
name="f_geo")
def forward(self, x):
def forward(self, x, targets=None):
f_det = self.det_conv1(x)
f_det = self.det_conv2(f_det)
f_score = self.score_conv(f_det)
......
ppocr/modeling/heads/det_sast_head.py
浏览文件 @ 253b8453
......@@ -116,7 +116,7 @@ class SASTHead(nn.Layer):
self.head1 = SAST_Header1(in_channels)
self.head2 = SAST_Header2(in_channels)
def forward(self, x):
def forward(self, x, targets=None):
f_score, f_border = self.head1(x)
f_tvo, f_tco = self.head2(x)
......
ppocr/modeling/heads/e2e_pg_head.py
浏览文件 @ 253b8453
......@@ -220,7 +220,7 @@ class PGHead(nn.Layer):
weight_attr=ParamAttr(name="conv_f_direc{}".format(4)),
bias_attr=False)
def forward(self, x):
def forward(self, x, targets=None):
f_score = self.conv_f_score1(x)
f_score = self.conv_f_score2(f_score)
f_score = self.conv_f_score3(f_score)
......
ppocr/modeling/heads/rec_ctc_head.py
浏览文件 @ 253b8453
......@@ -23,32 +23,57 @@ from paddle import ParamAttr, nn
from paddle.nn import functional as F
def get_para_bias_attr(l2_decay, k, name):
def get_para_bias_attr(l2_decay, k):
regularizer = paddle.regularizer.L2Decay(l2_decay)
stdv = 1.0 / math.sqrt(k * 1.0)
initializer = nn.initializer.Uniform(-stdv, stdv)
weight_attr = ParamAttr(
regularizer=regularizer, initializer=initializer, name=name + "_w_attr")
bias_attr = ParamAttr(
regularizer=regularizer, initializer=initializer, name=name + "_b_attr")
weight_attr = ParamAttr(regularizer=regularizer, initializer=initializer)
bias_attr = ParamAttr(regularizer=regularizer, initializer=initializer)
return [weight_attr, bias_attr]
class CTCHead(nn.Layer):
def __init__(self, in_channels, out_channels, fc_decay=0.0004, **kwargs):
def __init__(self,
in_channels,
out_channels,
fc_decay=0.0004,
mid_channels=None,
**kwargs):
super(CTCHead, self).__init__()
if mid_channels is None:
weight_attr, bias_attr = get_para_bias_attr(
l2_decay=fc_decay, k=in_channels, name='ctc_fc')
l2_decay=fc_decay, k=in_channels)
self.fc = nn.Linear(
in_channels,
out_channels,
weight_attr=weight_attr,
bias_attr=bias_attr,
name='ctc_fc')
bias_attr=bias_attr)
else:
weight_attr1, bias_attr1 = get_para_bias_attr(
l2_decay=fc_decay, k=in_channels)
self.fc1 = nn.Linear(
in_channels,
mid_channels,
weight_attr=weight_attr1,
bias_attr=bias_attr1)
weight_attr2, bias_attr2 = get_para_bias_attr(
l2_decay=fc_decay, k=mid_channels)
self.fc2 = nn.Linear(
mid_channels,
out_channels,
weight_attr=weight_attr2,
bias_attr=bias_attr2)
self.out_channels = out_channels
self.mid_channels = mid_channels
def forward(self, x, labels=None):
def forward(self, x, targets=None):
if self.mid_channels is None:
predicts = self.fc(x)
else:
predicts = self.fc1(x)
predicts = self.fc2(predicts)
if not self.training:
predicts = F.softmax(predicts, axis=2)
return predicts
ppocr/modeling/heads/rec_srn_head.py
浏览文件 @ 253b8453
......@@ -250,7 +250,8 @@ class SRNHead(nn.Layer):
self.gsrm.wrap_encoder1.prepare_decoder.emb0 = self.gsrm.wrap_encoder0.prepare_decoder.emb0
def forward(self, inputs, others):
def forward(self, inputs, targets=None):
others = targets[-4:]
encoder_word_pos = others[0]
gsrm_word_pos = others[1]
gsrm_slf_attn_bias1 = others[2]
......
ppocr/modeling/heads/table_att_head.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
import numpy as np
class TableAttentionHead(nn.Layer):
def __init__(self, in_channels, hidden_size, loc_type, in_max_len=488, **kwargs):
super(TableAttentionHead, self).__init__()
self.input_size = in_channels[-1]
self.hidden_size = hidden_size
self.elem_num = 30
self.max_text_length = 100
self.max_elem_length = 500
self.max_cell_num = 500
self.structure_attention_cell = AttentionGRUCell(
self.input_size, hidden_size, self.elem_num, use_gru=False)
self.structure_generator = nn.Linear(hidden_size, self.elem_num)
self.loc_type = loc_type
self.in_max_len = in_max_len
if self.loc_type == 1:
self.loc_generator = nn.Linear(hidden_size, 4)
else:
if self.in_max_len == 640:
self.loc_fea_trans = nn.Linear(400, self.max_elem_length+1)
elif self.in_max_len == 800:
self.loc_fea_trans = nn.Linear(625, self.max_elem_length+1)
else:
self.loc_fea_trans = nn.Linear(256, self.max_elem_length+1)
self.loc_generator = nn.Linear(self.input_size + hidden_size, 4)
def _char_to_onehot(self, input_char, onehot_dim):
input_ont_hot = F.one_hot(input_char, onehot_dim)
return input_ont_hot
def forward(self, inputs, targets=None):
# if and else branch are both needed when you want to assign a variable
# if you modify the var in just one branch, then the modification will not work.
fea = inputs[-1]
if len(fea.shape) == 3:
pass
else:
last_shape = int(np.prod(fea.shape[2:])) # gry added
fea = paddle.reshape(fea, [fea.shape[0], fea.shape[1], last_shape])
fea = fea.transpose([0, 2, 1]) # (NTC)(batch, width, channels)
batch_size = fea.shape[0]
hidden = paddle.zeros((batch_size, self.hidden_size))
output_hiddens = []
if self.training and targets is not None:
structure = targets[0]
for i in range(self.max_elem_length+1):
elem_onehots = self._char_to_onehot(
structure[:, i], onehot_dim=self.elem_num)
(outputs, hidden), alpha = self.structure_attention_cell(
hidden, fea, elem_onehots)
output_hiddens.append(paddle.unsqueeze(outputs, axis=1))
output = paddle.concat(output_hiddens, axis=1)
structure_probs = self.structure_generator(output)
if self.loc_type == 1:
loc_preds = self.loc_generator(output)
loc_preds = F.sigmoid(loc_preds)
else:
loc_fea = fea.transpose([0, 2, 1])
loc_fea = self.loc_fea_trans(loc_fea)
loc_fea = loc_fea.transpose([0, 2, 1])
loc_concat = paddle.concat([output, loc_fea], axis=2)
loc_preds = self.loc_generator(loc_concat)
loc_preds = F.sigmoid(loc_preds)
else:
temp_elem = paddle.zeros(shape=[batch_size], dtype="int32")
structure_probs = None
loc_preds = None
elem_onehots = None
outputs = None
alpha = None
max_elem_length = paddle.to_tensor(self.max_elem_length)
i = 0
while i < max_elem_length+1:
elem_onehots = self._char_to_onehot(
temp_elem, onehot_dim=self.elem_num)
(outputs, hidden), alpha = self.structure_attention_cell(
hidden, fea, elem_onehots)
output_hiddens.append(paddle.unsqueeze(outputs, axis=1))
structure_probs_step = self.structure_generator(outputs)
temp_elem = structure_probs_step.argmax(axis=1, dtype="int32")
i += 1
output = paddle.concat(output_hiddens, axis=1)
structure_probs = self.structure_generator(output)
structure_probs = F.softmax(structure_probs)
if self.loc_type == 1:
loc_preds = self.loc_generator(output)
loc_preds = F.sigmoid(loc_preds)
else:
loc_fea = fea.transpose([0, 2, 1])
loc_fea = self.loc_fea_trans(loc_fea)
loc_fea = loc_fea.transpose([0, 2, 1])
loc_concat = paddle.concat([output, loc_fea], axis=2)
loc_preds = self.loc_generator(loc_concat)
loc_preds = F.sigmoid(loc_preds)
return {'structure_probs':structure_probs, 'loc_preds':loc_preds}
class AttentionGRUCell(nn.Layer):
def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
super(AttentionGRUCell, self).__init__()
self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
self.h2h = nn.Linear(hidden_size, hidden_size)
self.score = nn.Linear(hidden_size, 1, bias_attr=False)
self.rnn = nn.GRUCell(
input_size=input_size + num_embeddings, hidden_size=hidden_size)
self.hidden_size = hidden_size
def forward(self, prev_hidden, batch_H, char_onehots):
batch_H_proj = self.i2h(batch_H)
prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden), axis=1)
res = paddle.add(batch_H_proj, prev_hidden_proj)
res = paddle.tanh(res)
e = self.score(res)
alpha = F.softmax(e, axis=1)
alpha = paddle.transpose(alpha, [0, 2, 1])
context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
concat_context = paddle.concat([context, char_onehots], 1)
cur_hidden = self.rnn(concat_context, prev_hidden)
return cur_hidden, alpha
class AttentionLSTM(nn.Layer):
def __init__(self, in_channels, out_channels, hidden_size, **kwargs):
super(AttentionLSTM, self).__init__()
self.input_size = in_channels
self.hidden_size = hidden_size
self.num_classes = out_channels
self.attention_cell = AttentionLSTMCell(
in_channels, hidden_size, out_channels, use_gru=False)
self.generator = nn.Linear(hidden_size, out_channels)
def _char_to_onehot(self, input_char, onehot_dim):
input_ont_hot = F.one_hot(input_char, onehot_dim)
return input_ont_hot
def forward(self, inputs, targets=None, batch_max_length=25):
batch_size = inputs.shape[0]
num_steps = batch_max_length
hidden = (paddle.zeros((batch_size, self.hidden_size)), paddle.zeros(
(batch_size, self.hidden_size)))
output_hiddens = []
if targets is not None:
for i in range(num_steps):
# one-hot vectors for a i-th char
char_onehots = self._char_to_onehot(
targets[:, i], onehot_dim=self.num_classes)
hidden, alpha = self.attention_cell(hidden, inputs,
char_onehots)
hidden = (hidden[1][0], hidden[1][1])
output_hiddens.append(paddle.unsqueeze(hidden[0], axis=1))
output = paddle.concat(output_hiddens, axis=1)
probs = self.generator(output)
else:
targets = paddle.zeros(shape=[batch_size], dtype="int32")
probs = None
for i in range(num_steps):
char_onehots = self._char_to_onehot(
targets, onehot_dim=self.num_classes)
hidden, alpha = self.attention_cell(hidden, inputs,
char_onehots)
probs_step = self.generator(hidden[0])
hidden = (hidden[1][0], hidden[1][1])
if probs is None:
probs = paddle.unsqueeze(probs_step, axis=1)
else:
probs = paddle.concat(
[probs, paddle.unsqueeze(
probs_step, axis=1)], axis=1)
next_input = probs_step.argmax(axis=1)
targets = next_input
return probs
class AttentionLSTMCell(nn.Layer):
def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
super(AttentionLSTMCell, self).__init__()
self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
self.h2h = nn.Linear(hidden_size, hidden_size)
self.score = nn.Linear(hidden_size, 1, bias_attr=False)
if not use_gru:
self.rnn = nn.LSTMCell(
input_size=input_size + num_embeddings, hidden_size=hidden_size)
else:
self.rnn = nn.GRUCell(
input_size=input_size + num_embeddings, hidden_size=hidden_size)
self.hidden_size = hidden_size
def forward(self, prev_hidden, batch_H, char_onehots):
batch_H_proj = self.i2h(batch_H)
prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden[0]), axis=1)
res = paddle.add(batch_H_proj, prev_hidden_proj)
res = paddle.tanh(res)
e = self.score(res)
alpha = F.softmax(e, axis=1)
alpha = paddle.transpose(alpha, [0, 2, 1])
context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
concat_context = paddle.concat([context, char_onehots], 1)
cur_hidden = self.rnn(concat_context, prev_hidden)
return cur_hidden, alpha
ppocr/modeling/necks/__init__.py
浏览文件 @ 253b8453
......@@ -21,7 +21,8 @@ def build_neck(config):
from .sast_fpn import SASTFPN
from .rnn import SequenceEncoder
from .pg_fpn import PGFPN
support_dict = ['DBFPN', 'EASTFPN', 'SASTFPN', 'SequenceEncoder', 'PGFPN']
from .table_fpn import TableFPN
support_dict = ['DBFPN', 'EASTFPN', 'SASTFPN', 'SequenceEncoder', 'PGFPN', 'TableFPN']
module_name = config.pop('name')
assert module_name in support_dict, Exception('neck only support {}'.format(
......
ppocr/modeling/necks/db_fpn.py
浏览文件 @ 253b8453
......@@ -32,61 +32,53 @@ class DBFPN(nn.Layer):
in_channels=in_channels[0],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(
name='conv2d_51.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.in3_conv = nn.Conv2D(
in_channels=in_channels[1],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(
name='conv2d_50.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.in4_conv = nn.Conv2D(
in_channels=in_channels[2],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(
name='conv2d_49.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.in5_conv = nn.Conv2D(
in_channels=in_channels[3],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(
name='conv2d_48.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p5_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(
name='conv2d_52.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p4_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(
name='conv2d_53.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p3_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(
name='conv2d_54.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p2_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(
name='conv2d_55.w_0', initializer=weight_attr),
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
def forward(self, x):
......
ppocr/modeling/necks/table_fpn.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import paddle
from paddle import nn
import paddle.nn.functional as F
from paddle import ParamAttr
class TableFPN(nn.Layer):
def __init__(self, in_channels, out_channels, **kwargs):
super(TableFPN, self).__init__()
self.out_channels = 512
weight_attr = paddle.nn.initializer.KaimingUniform()
self.in2_conv = nn.Conv2D(
in_channels=in_channels[0],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.in3_conv = nn.Conv2D(
in_channels=in_channels[1],
out_channels=self.out_channels,
kernel_size=1,
stride = 1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.in4_conv = nn.Conv2D(
in_channels=in_channels[2],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.in5_conv = nn.Conv2D(
in_channels=in_channels[3],
out_channels=self.out_channels,
kernel_size=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p5_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p4_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p3_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.p2_conv = nn.Conv2D(
in_channels=self.out_channels,
out_channels=self.out_channels // 4,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(initializer=weight_attr),
bias_attr=False)
self.fuse_conv = nn.Conv2D(
in_channels=self.out_channels * 4,
out_channels=512,
kernel_size=3,
padding=1,
weight_attr=ParamAttr(initializer=weight_attr), bias_attr=False)
def forward(self, x):
c2, c3, c4, c5 = x
in5 = self.in5_conv(c5)
in4 = self.in4_conv(c4)
in3 = self.in3_conv(c3)
in2 = self.in2_conv(c2)
out4 = in4 + F.upsample(
in5, size=in4.shape[2:4], mode="nearest", align_mode=1) # 1/16
out3 = in3 + F.upsample(
out4, size=in3.shape[2:4], mode="nearest", align_mode=1) # 1/8
out2 = in2 + F.upsample(
out3, size=in2.shape[2:4], mode="nearest", align_mode=1) # 1/4
p4 = F.upsample(out4, size=in5.shape[2:4], mode="nearest", align_mode=1)
p3 = F.upsample(out3, size=in5.shape[2:4], mode="nearest", align_mode=1)
p2 = F.upsample(out2, size=in5.shape[2:4], mode="nearest", align_mode=1)
fuse = paddle.concat([in5, p4, p3, p2], axis=1)
fuse_conv = self.fuse_conv(fuse) * 0.005
return [c5 + fuse_conv]
ppocr/postprocess/__init__.py
浏览文件 @ 253b8453
......@@ -21,18 +21,20 @@ import copy
__all__ = ['build_post_process']
from .db_postprocess import DBPostProcess
from .east_postprocess import EASTPostProcess
from .sast_postprocess import SASTPostProcess
from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode, DistillationCTCLabelDecode, \
TableLabelDecode
from .cls_postprocess import ClsPostProcess
from .pg_postprocess import PGPostProcess
def build_post_process(config, global_config=None):
from .db_postprocess import DBPostProcess
from .east_postprocess import EASTPostProcess
from .sast_postprocess import SASTPostProcess
from .rec_postprocess import CTCLabelDecode, AttnLabelDecode, SRNLabelDecode
from .cls_postprocess import ClsPostProcess
from .pg_postprocess import PGPostProcess
def build_post_process(config, global_config=None):
support_dict = [
'DBPostProcess', 'EASTPostProcess', 'SASTPostProcess', 'CTCLabelDecode',
'AttnLabelDecode', 'ClsPostProcess', 'SRNLabelDecode', 'PGPostProcess'
'AttnLabelDecode', 'ClsPostProcess', 'SRNLabelDecode', 'PGPostProcess',
'DistillationCTCLabelDecode', 'TableLabelDecode'
]
config = copy.deepcopy(config)
......
ppocr/postprocess/rec_postprocess.py
浏览文件 @ 253b8453
......@@ -44,16 +44,16 @@ class BaseRecLabelDecode(object):
self.character_str = string.printable[:-6]
dict_character = list(self.character_str)
elif character_type in support_character_type:
self.character_str = ""
self.character_str = []
assert character_dict_path is not None, "character_dict_path should not be None when character_type is {}".format(
character_type)
with open(character_dict_path, "rb") as fin:
lines = fin.readlines()
for line in lines:
line = line.decode('utf-8').strip("\n").strip("\r\n")
self.character_str += line
self.character_str.append(line)
if use_space_char:
self.character_str += " "
self.character_str.append(" ")
dict_character = list(self.character_str)
else:
......@@ -125,6 +125,37 @@ class CTCLabelDecode(BaseRecLabelDecode):
return dict_character
class DistillationCTCLabelDecode(CTCLabelDecode):
"""
Convert
Convert between text-label and text-index
"""
def __init__(self,
character_dict_path=None,
character_type='ch',
use_space_char=False,
model_name=["student"],
key=None,
**kwargs):
super(DistillationCTCLabelDecode, self).__init__(
character_dict_path, character_type, use_space_char)
if not isinstance(model_name, list):
model_name = [model_name]
self.model_name = model_name
self.key = key
def __call__(self, preds, label=None, *args, **kwargs):
output = dict()
for name in self.model_name:
pred = preds[name]
if self.key is not None:
pred = pred[self.key]
output[name] = super().__call__(pred, label=label, *args, **kwargs)
return output
class AttnLabelDecode(BaseRecLabelDecode):
""" Convert between text-label and text-index """
......@@ -288,3 +319,138 @@ class SRNLabelDecode(BaseRecLabelDecode):
assert False, "unsupport type %s in get_beg_end_flag_idx" \
% beg_or_end
return idx
class TableLabelDecode(object):
""" """
def __init__(self,
character_dict_path,
**kwargs):
list_character, list_elem = self.load_char_elem_dict(character_dict_path)
list_character = self.add_special_char(list_character)
list_elem = self.add_special_char(list_elem)
self.dict_character = {}
self.dict_idx_character = {}
for i, char in enumerate(list_character):
self.dict_idx_character[i] = char
self.dict_character[char] = i
self.dict_elem = {}
self.dict_idx_elem = {}
for i, elem in enumerate(list_elem):
self.dict_idx_elem[i] = elem
self.dict_elem[elem] = i
def load_char_elem_dict(self, character_dict_path):
list_character = []
list_elem = []
with open(character_dict_path, "rb") as fin:
lines = fin.readlines()
substr = lines[0].decode('utf-8').strip("\n").split("\t")
character_num = int(substr[0])
elem_num = int(substr[1])
for cno in range(1, 1 + character_num):
character = lines[cno].decode('utf-8').strip("\n")
list_character.append(character)
for eno in range(1 + character_num, 1 + character_num + elem_num):
elem = lines[eno].decode('utf-8').strip("\n")
list_elem.append(elem)
return list_character, list_elem
def add_special_char(self, list_character):
self.beg_str = "sos"
self.end_str = "eos"
list_character = [self.beg_str] + list_character + [self.end_str]
return list_character
def __call__(self, preds):
structure_probs = preds['structure_probs']
loc_preds = preds['loc_preds']
if isinstance(structure_probs,paddle.Tensor):
structure_probs = structure_probs.numpy()
if isinstance(loc_preds,paddle.Tensor):
loc_preds = loc_preds.numpy()
structure_idx = structure_probs.argmax(axis=2)
structure_probs = structure_probs.max(axis=2)
structure_str, structure_pos, result_score_list, result_elem_idx_list = self.decode(structure_idx,
structure_probs, 'elem')
res_html_code_list = []
res_loc_list = []
batch_num = len(structure_str)
for bno in range(batch_num):
res_loc = []
for sno in range(len(structure_str[bno])):
text = structure_str[bno][sno]
if text in ['<td>', '<td']:
pos = structure_pos[bno][sno]
res_loc.append(loc_preds[bno, pos])
res_html_code = ''.join(structure_str[bno])
res_loc = np.array(res_loc)
res_html_code_list.append(res_html_code)
res_loc_list.append(res_loc)
return {'res_html_code': res_html_code_list, 'res_loc': res_loc_list, 'res_score_list': result_score_list,
'res_elem_idx_list': result_elem_idx_list,'structure_str_list':structure_str}
def decode(self, text_index, structure_probs, char_or_elem):
"""convert text-label into text-index.
"""
if char_or_elem == "char":
current_dict = self.dict_idx_character
else:
current_dict = self.dict_idx_elem
ignored_tokens = self.get_ignored_tokens('elem')
beg_idx, end_idx = ignored_tokens
result_list = []
result_pos_list = []
result_score_list = []
result_elem_idx_list = []
batch_size = len(text_index)
for batch_idx in range(batch_size):
char_list = []
elem_pos_list = []
elem_idx_list = []
score_list = []
for idx in range(len(text_index[batch_idx])):
tmp_elem_idx = int(text_index[batch_idx][idx])
if idx > 0 and tmp_elem_idx == end_idx:
break
if tmp_elem_idx in ignored_tokens:
continue
char_list.append(current_dict[tmp_elem_idx])
elem_pos_list.append(idx)
score_list.append(structure_probs[batch_idx, idx])
elem_idx_list.append(tmp_elem_idx)
result_list.append(char_list)
result_pos_list.append(elem_pos_list)
result_score_list.append(score_list)
result_elem_idx_list.append(elem_idx_list)
return result_list, result_pos_list, result_score_list, result_elem_idx_list
def get_ignored_tokens(self, char_or_elem):
beg_idx = self.get_beg_end_flag_idx("beg", char_or_elem)
end_idx = self.get_beg_end_flag_idx("end", char_or_elem)
return [beg_idx, end_idx]
def get_beg_end_flag_idx(self, beg_or_end, char_or_elem):
if char_or_elem == "char":
if beg_or_end == "beg":
idx = self.dict_character[self.beg_str]
elif beg_or_end == "end":
idx = self.dict_character[self.end_str]
else:
assert False, "Unsupport type %s in get_beg_end_flag_idx of char" \
% beg_or_end
elif char_or_elem == "elem":
if beg_or_end == "beg":
idx = self.dict_elem[self.beg_str]
elif beg_or_end == "end":
idx = self.dict_elem[self.end_str]
else:
assert False, "Unsupport type %s in get_beg_end_flag_idx of elem" \
% beg_or_end
else:
assert False, "Unsupport type %s in char_or_elem" \
% char_or_elem
return idx
ppocr/utils/dict/table_dict.txt 0 → 100644
浏览文件 @ 253b8453
←
</overline>
☆
─
α

⋅
$
ω
ψ
χ
(
υ
≥
σ
,
ρ
ε
0
■
4
8
✗
b
<
✓
Ψ
Ω
€
D
3
Π
H
║
</strike>
L
Φ
Χ
θ
P
κ
λ
μ
T
ξ
X
β
γ
δ
\
ζ
η
`
d
<strike>
h
f
l
Θ
p
√
t
</sub>
x
Β
Γ
Δ
|
ǂ
ɛ
j
̧
➢
⁡
̌
′
«
△
▲
#
</b>
'
Ι
+
/
▼
⇑
□
·
7
▪
;
?
➔
∩
C
÷
G
⇒
K
<sup>
O
S
С
W
Α
[
○
_
●
‡
c
z
g
<i>
o
<sub>
〈
〉
s
⩽
w
φ
ʹ
{
»
∣
̆
e
ˆ
∈
τ
◆
ι
∅
∆
∙
∘
Ø
ß
✔
∞
∑
−
×
◊
∗
∖
˃
˂
∫
"
i
&
π
↔
*
∥
æ
∧
.
⁄
ø
Q
∼
6
⁎
:
★
>
a
B
≈
F
J
̄
N
♯
R
V
<overline>
―
Z
♣
^
¤
¥
§
<underline>
¢
£
≦
­
≤
‖
Λ
©
n
↓
→
↑
r
°
±
v
<b>
♂
k
♀
~
ᅟ
̇
@
”
♦
ł
®
⊕
„
!
</sup>
%
⇓
)
-
1
5
9
=
А
A
‰
⋆
Σ
E
◦
I
※
M
m
̨
⩾
†
</i>
•
U
Y


]
̸
2
‐
–
‒
̂
—
̀
́
’
‘
⋮
⋯
̊
“
̈
≧
q
u
ı
y
</underline>
​
̃
}
ν
ppocr/utils/dict/table_structure_dict.txt 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
ppocr/utils/logging.py
浏览文件 @ 253b8453
......@@ -22,7 +22,7 @@ logger_initialized = {}
@functools.lru_cache()
def get_logger(name='root', log_file=None, log_level=logging.INFO):
def get_logger(name='root', log_file=None, log_level=logging.DEBUG):
"""Initialize and get a logger by name.
If the logger has not been initialized, this method will initialize the
logger by adding one or two handlers, otherwise the initialized logger will
......
ppocr/utils/network.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import sys
import tarfile
import requests
from tqdm import tqdm
from ppocr.utils.logging import get_logger
def download_with_progressbar(url, save_path):
logger = get_logger()
response = requests.get(url, stream=True)
total_size_in_bytes = int(response.headers.get('content-length', 0))
block_size = 1024 # 1 Kibibyte
progress_bar = tqdm(total=total_size_in_bytes, unit='iB', unit_scale=True)
with open(save_path, 'wb') as file:
for data in response.iter_content(block_size):
progress_bar.update(len(data))
file.write(data)
progress_bar.close()
if total_size_in_bytes == 0 or progress_bar.n != total_size_in_bytes:
logger.error("Something went wrong while downloading models")
sys.exit(0)
def maybe_download(model_storage_directory, url):
# using custom model
tar_file_name_list = [
'inference.pdiparams', 'inference.pdiparams.info', 'inference.pdmodel'
]
if not os.path.exists(
os.path.join(model_storage_directory, 'inference.pdiparams')
) or not os.path.exists(
os.path.join(model_storage_directory, 'inference.pdmodel')):
assert url.endswith('.tar'), 'Only supports tar compressed package'
tmp_path = os.path.join(model_storage_directory, url.split('/')[-1])
print('download {} to {}'.format(url, tmp_path))
os.makedirs(model_storage_directory, exist_ok=True)
download_with_progressbar(url, tmp_path)
with tarfile.open(tmp_path, 'r') as tarObj:
for member in tarObj.getmembers():
filename = None
for tar_file_name in tar_file_name_list:
if tar_file_name in member.name:
filename = tar_file_name
if filename is None:
continue
file = tarObj.extractfile(member)
with open(
os.path.join(model_storage_directory, filename),
'wb') as f:
f.write(file.read())
os.remove(tmp_path)
def is_link(s):
return s is not None and s.startswith('http')
def confirm_model_dir_url(model_dir, default_model_dir, default_url):
url = default_url
if model_dir is None or is_link(model_dir):
if is_link(model_dir):
url = model_dir
file_name = url.split('/')[-1][:-4]
model_dir = default_model_dir
model_dir = os.path.join(model_dir, file_name)
return model_dir, url
ppocr/utils/save_load.py
浏览文件 @ 253b8453
......@@ -23,6 +23,8 @@ import six
import paddle
from ppocr.utils.logging import get_logger
__all__ = ['init_model', 'save_model', 'load_dygraph_pretrain']
......@@ -42,44 +44,11 @@ def _mkdir_if_not_exist(path, logger):
raise OSError('Failed to mkdir {}'.format(path))
def load_dygraph_pretrain(model, logger, path=None, load_static_weights=False):
if not (os.path.isdir(path) or os.path.exists(path + '.pdparams')):
raise ValueError("Model pretrain path {} does not "
"exists.".format(path))
if load_static_weights:
pre_state_dict = paddle.static.load_program_state(path)
param_state_dict = {}
model_dict = model.state_dict()
for key in model_dict.keys():
weight_name = model_dict[key].name
weight_name = weight_name.replace('binarize', '').replace(
'thresh', '') # for DB
if weight_name in pre_state_dict.keys():
# logger.info('Load weight: {}, shape: {}'.format(
# weight_name, pre_state_dict[weight_name].shape))
if 'encoder_rnn' in key:
# delete axis which is 1
pre_state_dict[weight_name] = pre_state_dict[
weight_name].squeeze()
# change axis
if len(pre_state_dict[weight_name].shape) > 1:
pre_state_dict[weight_name] = pre_state_dict[
weight_name].transpose((1, 0))
param_state_dict[key] = pre_state_dict[weight_name]
else:
param_state_dict[key] = model_dict[key]
model.set_state_dict(param_state_dict)
return
param_state_dict = paddle.load(path + '.pdparams')
model.set_state_dict(param_state_dict)
return
def init_model(config, model, logger, optimizer=None, lr_scheduler=None):
def init_model(config, model, optimizer=None, lr_scheduler=None):
"""
load model from checkpoint or pretrained_model
"""
logger = get_logger()
global_config = config['Global']
checkpoints = global_config.get('checkpoints')
pretrained_model = global_config.get('pretrained_model')
......@@ -102,18 +71,17 @@ def init_model(config, model, logger, optimizer=None, lr_scheduler=None):
best_model_dict = states_dict.get('best_model_dict', {})
if 'epoch' in states_dict:
best_model_dict['start_epoch'] = states_dict['epoch'] + 1
logger.info("resume from {}".format(checkpoints))
elif pretrained_model:
load_static_weights = global_config.get('load_static_weights', False)
if not isinstance(pretrained_model, list):
pretrained_model = [pretrained_model]
if not isinstance(load_static_weights, list):
load_static_weights = [load_static_weights] * len(pretrained_model)
for idx, pretrained in enumerate(pretrained_model):
load_static = load_static_weights[idx]
load_dygraph_pretrain(
model, logger, path=pretrained, load_static_weights=load_static)
for pretrained in pretrained_model:
if not (os.path.isdir(pretrained) or
os.path.exists(pretrained + '.pdparams')):
raise ValueError("Model pretrain path {} does not "
"exists.".format(pretrained))
param_state_dict = paddle.load(pretrained + '.pdparams')
model.set_state_dict(param_state_dict)
logger.info("load pretrained model from {}".format(
pretrained_model))
else:
......
test1/MANIFEST.in 0 → 100644
浏览文件 @ 253b8453
include LICENSE
include README.md
recursive-include ppocr/utils *.txt utility.py logging.py network.py
recursive-include ppocr/data/ *.py
recursive-include ppocr/postprocess *.py
recursive-include tools/infer *.py
recursive-include ppstructure *.py
test1/__init__.py 0 → 100644
浏览文件 @ 253b8453
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from .paddlestructure import PaddleStructure, draw_result, to_excel
__all__ = ['PaddleStructure', 'draw_result', 'to_excel']
test1/api.md 0 → 100644
浏览文件 @ 253b8453
# PaddleStructure
install layoutparser
```sh
wget https://paddleocr.bj.bcebos.com/whl/layoutparser-0.0.0-py3-none-any.whl
pip3 install layoutparser-0.0.0-py3-none-any.whl
```
## 1. Introduction to pipeline
PaddleStructure is a toolkit for complex layout text OCR, the process is as follows
![pipeline](../doc/table/pipeline.png)
In PaddleStructure, the image will be analyzed by layoutparser first. In the layout analysis, the area in the image will be classified, and the OCR process will be carried out according to the category.
Currently layoutparser will output five categories:
1. Text
2. Title
3. Figure
4. List
5. Table
Types 1-4 follow the traditional OCR process, and 5 follow the Table OCR process.
## 2. LayoutParser
## 3. Table OCR
[doc](table/README.md)
## 4. Predictive by inference engine
Use the following commands to complete the inference
```python
python3 table/predict_system.py --det_model_dir=path/to/det_model_dir --rec_model_dir=path/to/rec_model_dir --table_model_dir=path/to/table_model_dir --image_dir=../doc/table/1.png --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt --rec_char_type=EN --det_limit_side_len=736 --det_limit_type=min --output ../output/table
```
After running, each image will have a directory with the same name under the directory specified in the output field. Each table in the picture will be stored as an excel, and the excel file name will be the coordinates of the table in the image.
## 5. PaddleStructure whl package introduction
### 5.1 Use
5.1.1 Use by code
```python
import os
import cv2
from paddlestructure import PaddleStructure,draw_result,save_res
table_engine = PaddleStructure(show_log=True)
save_folder = './output/table'
img_path = '../doc/table/1.png'
img = cv2.imread(img_path)
result = table_engine(img)
save_res(result, save_folder,os.path.basename(img_path).split('.')[0])
for line in result:
print(line)
from PIL import Image
font_path = 'path/tp/PaddleOCR/doc/fonts/simfang.ttf'
image = Image.open(img_path).convert('RGB')
im_show = draw_result(image, result,font_path=font_path)
im_show = Image.fromarray(im_show)
im_show.save('result.jpg')
```
5.1.2 Use by command line
```bash
paddlestructure --image_dir=../doc/table/1.png
```
### Parameter Description
Most of the parameters are consistent with the paddleocr whl package, see [whl package documentation](../doc/doc_ch/whl.md)
| Parameter | Description | Default |
|------------------------|------------------------------------------------------|------------------|
| output | The path where excel and recognition results are saved | ./output/table |
| structure_max_len | When the table structure model predicts, the long side of the image is resized | 488 |
| structure_model_dir | Table structure inference model path | None |
| structure_char_type | Dictionary path used by table structure model | ../ppocr/utils/dict/table_structure_dict.tx |
test1/api_ch.md 0 → 100644
浏览文件 @ 253b8453
# PaddleStructure
安装layoutparser
```sh
wget https://paddleocr.bj.bcebos.com/whl/layoutparser-0.0.0-py3-none-any.whl
pip3 install layoutparser-0.0.0-py3-none-any.whl
```
## 1. pipeline介绍
PaddleStructure 是一个用于复杂板式文字OCR的工具包，流程如下
![pipeline](../doc/table/pipeline.png)
在PaddleStructure中，图片会先经由layoutparser进行版面分析，在版面分析中，会对图片里的区域进行分类，根据根据类别进行对于的ocr流程。
目前layoutparser会输出五个类别:
1. Text
2. Title
3. Figure
4. List
5. Table
1-4类走传统的OCR流程，5走表格的OCR流程。
## 2. LayoutParser
[文档](layout/README.md)
## 3. Table OCR
[文档](table/README_ch.md)
## 4. 预测引擎推理
使用如下命令即可完成预测引擎的推理
```python
python3 table/predict_system.py --det_model_dir=path/to/det_model_dir --rec_model_dir=path/to/rec_model_dir --table_model_dir=path/to/table_model_dir --image_dir=../doc/table/1.png --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt --rec_char_type=EN --det_limit_side_len=736 --det_limit_type=min --output ../output/table
```
运行完成后，每张图片会output字段指定的目录下有一个同名目录，图片里的每个表格会存储为一个excel，excel文件名为表格在图片里的坐标。
## 5. PaddleStructure whl包介绍
### 5.1 使用
5.1.1 代码使用
```python
import os
import cv2
from paddlestructure import PaddleStructure,draw_result,save_res
table_engine = PaddleStructure(show_log=True)
save_folder = './output/table'
img_path = '../doc/table/1.png'
img = cv2.imread(img_path)
result = table_engine(img)
save_res(result, save_folder,os.path.basename(img_path).split('.')[0])
for line in result:
print(line)
from PIL import Image
font_path = 'path/tp/PaddleOCR/doc/fonts/simfang.ttf'
image = Image.open(img_path).convert('RGB')
im_show = draw_result(image, result,font_path=font_path)
im_show = Image.fromarray(im_show)
im_show.save('result.jpg')
```
5.1.2 命令行使用
```bash
paddlestructure --image_dir=../doc/table/1.png
```
### 参数说明
大部分参数和paddleocr whl包保持一致，见 [whl包文档](../doc/doc_ch/whl.md)
| 字段 | 说明 | 默认值 |
|------------------------|------------------------------------------------------|------------------|
| output | excel和识别结果保存的地址 | ./output/table |
| table_max_len | 表格结构模型预测时，图像的长边resize尺度 | 488 |
| table_model_dir | 表格结构模型 inference 模型地址 | None |
| table_char_type | 表格结构模型所用字典地址 | ../ppocr/utils/dict/table_structure_dict.tx |
test1/layout/README.md 0 → 100644
浏览文件 @ 253b8453
# 版面分析使用说明
* [1. 安装whl包](#安装whl包)
* [2. 使用](#使用)
* [3. 后处理](#后处理)
* [4. 指标](#指标)
* [5. 训练版面分析模型](#训练版面分析模型)
<a name="安装whl包"></a>
## 1. 安装whl包
```bash
wget https://paddleocr.bj.bcebos.com/whl/layoutparser-0.0.0-py3-none-any.whl
pip install -U layoutparser-0.0.0-py3-none-any.whl
```
<a name="使用"></a>
## 2. 使用
使用layoutparser识别给定文档的布局：
```python
import layoutparser as lp
image = cv2.imread("imags/paper-image.jpg")
image = image[..., ::-1]
# 加载模型
model = lp.PaddleDetectionLayoutModel(config_path="lp://PubLayNet/ppyolov2_r50vd_dcn_365e_publaynet/config",
threshold=0.5,
label_map={0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"},
enforce_cpu=False,
enable_mkldnn=True)
# 检测
layout = model.detect(image)
# 显示结果
lp.draw_box(image, layout, box_width=3, show_element_type=True)
```
下图展示了结果，不同颜色的检测框表示不同的类别，并通过`show_element_type`在框的左上角显示具体类别：
<div align="center">
<img src="../../doc/table/result_all.jpg" width = "600" />
</div>
`PaddleDetectionLayoutModel`函数参数说明如下:
| 参数 | 含义 | 默认值 | 备注 |
| :------------: | :-------------------------: | :---------: | :----------------------------------------------------------: |
| config_path | 模型配置路径 | None | 指定config_path会自动下载模型(仅第一次，之后模型存在，不会再下载) |
| model_path | 模型路径 | None | 本地模型路径，config_path和model_path必须设置一个，不能同时为None |
| threshold | 预测得分的阈值 | 0.5 | \ |
| input_shape | reshape之后图片尺寸 | [3,640,640] | \ |
| batch_size | 测试batch size | 1 | \ |
| label_map | 类别映射表 | None | 设置config_path时，可以为None，根据数据集名称自动获取label_map |
| enforce_cpu | 代码是否使用CPU运行 | False | 设置为False表示使用GPU，True表示强制使用CPU |
| enforce_mkldnn | CPU预测中是否开启MKLDNN加速 | True | \ |
| thread_num | 设置CPU线程数 | 10 | \ |
目前支持以下几种模型配置和label map，您可以通过修改 `--config_path`和 `--label_map`使用这些模型，从而检测不同类型的内容：
| dataset | config_path | label_map |
| ------------------------------------------------------------ | ------------------------------------------------------------ | --------------------------------------------------------- |
| [TableBank](https://doc-analysis.github.io/tablebank-page/index.html) word | lp://TableBank/ppyolov2_r50vd_dcn_365e_tableBank_word/config | {0:"Table"} |
| TableBank latex | lp://TableBank/ppyolov2_r50vd_dcn_365e_tableBank_latex/config | {0:"Table"} |
| [PubLayNet](https://github.com/ibm-aur-nlp/PubLayNet) | lp://PubLayNet/ppyolov2_r50vd_dcn_365e_publaynet/config | {0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"} |
* TableBank word和TableBank latex分别在word文档、latex文档数据集训练；
* 下载TableBank数据集同时包含word和latex。
<a name="后处理"></a>
## 3. 后处理
版面分析检测包含多个类别，如果只想获取指定类别(如"Text"类别)的检测框、可以使用下述代码：
```python
# 首先过滤特定文本类型的区域
text_blocks = lp.Layout([b for b in layout if b.type=='Text'])
figure_blocks = lp.Layout([b for b in layout if b.type=='Figure'])
# 因为在图像区域内可能检测到文本区域，所以只需要删除它们
text_blocks = lp.Layout([b for b in text_blocks \
if not any(b.is_in(b_fig) for b_fig in figure_blocks)])
# 对文本区域排序并分配id
h, w = image.shape[:2]
left_interval = lp.Interval(0, w/2*1.05, axis='x').put_on_canvas(image)
left_blocks = text_blocks.filter_by(left_interval, center=True)
left_blocks.sort(key = lambda b:b.coordinates[1])
right_blocks = [b for b in text_blocks if b not in left_blocks]
right_blocks.sort(key = lambda b:b.coordinates[1])
# 最终合并两个列表，并按顺序添加索引
text_blocks = lp.Layout([b.set(id = idx) for idx, b in enumerate(left_blocks + right_blocks)])
# 显示结果
lp.draw_box(image, text_blocks,
box_width=3,
show_element_id=True)
```
显示只有"Text"类别的结果：
<div align="center">
<img src="../../doc/table/result_text.jpg" width = "600" />
</div>
<a name="指标"></a>
## 4. 指标
| Dataset | mAP | CPU time cost | GPU time cost |
| --------- | ---- | ------------- | ------------- |
| PubLayNet | 93.6 | 1713.7ms | 66.6ms |
| TableBank | 96.2 | 1968.4ms | 65.1ms |
**Envrionment：**
​ **CPU：** Intel(R) Xeon(R) CPU E5-2650 v4 @ 2.20GHz，24core
​ **GPU：** a single NVIDIA Tesla P40
<a name="训练版面分析模型"></a>
## 5. 训练版面分析模型
上述模型基于[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection) 训练，如果您想训练自己的版面分析模型，请参考：[train_layoutparser_model](train_layoutparser_model.md)
test1/layout/train_layoutparser_model.md 0 → 100644
浏览文件 @ 253b8453
# 训练版面分析
* [1. 安装](#安装)
* [1.1 环境要求](#环境要求)
* [1.2 安装PaddleDetection](#安装PaddleDetection)
* [2. 准备数据](#准备数据)
* [3. 配置文件改动和说明](#配置文件改动和说明)
* [4. PaddleDetection训练](#训练)
* [5. PaddleDetection预测](#预测)
* [6. 预测部署](#预测部署)
* [6.1 模型导出](#模型导出)
* [6.2 layout parser预测](#layout_parser预测)
<a name="安装"></a>
## 1. 安装
<a name="环境要求"></a>
### 1.1 环境要求
- PaddlePaddle 2.1
- OS 64 bit
- Python 3(3.5.1+/3.6/3.7/3.8/3.9)，64 bit
- pip/pip3(9.0.1+), 64 bit
- CUDA >= 10.1
- cuDNN >= 7.6
<a name="安装PaddleDetection"></a>
### 1.2 安装PaddleDetection
```bash
# 克隆PaddleDetection仓库
cd <path/to/clone/PaddleDetection>
git clone https://github.com/PaddlePaddle/PaddleDetection.git
cd PaddleDetection
# 安装其他依赖
pip install -r requirements.txt
```
更多安装教程，请参考: [Install doc](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.1/docs/tutorials/INSTALL_cn.md)
<a name="数据准备"></a>
## 2. 准备数据
下载 [PubLayNet](https://github.com/ibm-aur-nlp/PubLayNet) 数据集：
```bash
cd PaddleDetection/dataset/
mkdir publaynet
# 执行命令，下载
wget -O publaynet.tar.gz https://dax-cdn.cdn.appdomain.cloud/dax-publaynet/1.0.0/publaynet.tar.gz?_ga=2.104193024.1076900768.1622560733-649911202.1622560733
# 解压
tar -xvf publaynet.tar.gz
```
解压之后PubLayNet目录结构：
| File or Folder | Description | num |
| :------------- | :----------------------------------------------- | ------- |
| `train/` | Images in the training subset | 335,703 |
| `val/` | Images in the validation subset | 11,245 |
| `test/` | Images in the testing subset | 11,405 |
| `train.json` | Annotations for training images | |
| `val.json` | Annotations for validation images | |
| `LICENSE.txt` | Plaintext version of the CDLA-Permissive license | |
| `README.txt` | Text file with the file names and description | |
如果使用其它数据集，请参考[准备训练数据](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.1/docs/tutorials/PrepareDataSet.md)
<a name="配置文件改动和说明"></a>
## 3. 配置文件改动和说明
我们使用 `configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml`配置进行训练，配置文件摘要如下：
<div align='center'>
<img src='../../doc/table/PaddleDetection_config.png' width='600px'/>
</div>
从上图看到 `ppyolov2_r50vd_dcn_365e_coco.yml` 配置需要依赖其他的配置文件，在该例子中需要依赖:
```
coco_detection.yml：主要说明了训练数据和验证数据的路径
runtime.yml：主要说明了公共的运行参数，比如是否使用GPU、每多少个epoch存储checkpoint等
optimizer_365e.yml：主要说明了学习率和优化器的配置
ppyolov2_r50vd_dcn.yml：主要说明模型和主干网络的情况
ppyolov2_reader.yml：主要说明数据读取器配置，如batch size，并发加载子进程数等，同时包含读取后预处理操作，如resize、数据增强等等
```
根据实际情况，修改上述文件，比如数据集路径、batch size等。
<a name="训练"></a>
## 4. PaddleDetection训练
PaddleDetection提供了单卡/多卡训练模式，满足用户多种训练需求
* GPU 单卡训练
```bash
export CUDA_VISIBLE_DEVICES=0 #windows和Mac下不需要执行该命令
python tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml
```
* GPU多卡训练
```bash
export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --eval
```
--eval：表示边训练边验证
* 模型恢复训练
在日常训练过程中，有的用户由于一些原因导致训练中断，用户可以使用-r的命令恢复训练:
```bash
export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --eval -r output/ppyolov2_r50vd_dcn_365e_coco/10000
```
注意：如果遇到 "`Out of memory error`" 问题, 尝试在 `ppyolov2_reader.yml` 文件中调小`batch_size`
<a name="预测"></a>
## 5. PaddleDetection预测
设置参数，使用PaddleDetection预测：
```bash
export CUDA_VISIBLE_DEVICES=0
python tools/infer.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --infer_img=images/paper-image.jpg --output_dir=infer_output/ --draw_threshold=0.5 -o weights=output/ppyolov2_r50vd_dcn_365e_coco/model_final --use_vdl=Ture
```
`--draw_threshold` 是个可选参数. 根据 [NMS](https://ieeexplore.ieee.org/document/1699659) 的计算，不同阈值会产生不同的结果 `keep_top_k`表示设置输出目标的最大数量，默认值为100，用户可以根据自己的实际情况进行设定。
<a name="预测部署"></a>
## 6. 预测部署
在layout parser中使用自己训练好的模型，
<a name="模型导出"></a>
### 6.1 模型导出
在模型训练过程中保存的模型文件是包含前向预测和反向传播的过程，在实际的工业部署则不需要反向传播，因此需要将模型进行导成部署需要的模型格式。 在PaddleDetection中提供了 `tools/export_model.py`脚本来导出模型。
导出模型名称默认是`model.*`，layout parser代码模型名称是`inference.*`, 所以修改[PaddleDetection/ppdet/engine/trainer.py ](https://github.com/PaddlePaddle/PaddleDetection/blob/b87a1ea86fa18ce69e44a17ad1b49c1326f19ff9/ppdet/engine/trainer.py#L512) (点开链接查看详细代码行)，将`model`改为`inference`即可。
执行导出模型脚本：
```bash
python tools/export_model.py -c configs/ppyolo/ppyolov2_r50vd_dcn_365e_coco.yml --output_dir=./inference -o weights=output/ppyolov2_r50vd_dcn_365e_coco/model_final.pdparams
```
预测模型会导出到`inference/ppyolov2_r50vd_dcn_365e_coco`目录下，分别为`infer_cfg.yml`(预测不需要), `inference.pdiparams`, `inference.pdiparams.info`,`inference.pdmodel` 。
更多模型导出教程，请参考：[EXPORT_MODEL](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.1/deploy/EXPORT_MODEL.md)
<a name="layout parser预测"></a>
### 6.2 layout_parser预测
`model_path`指定训练好的模型路径，使用layout parser进行预测：
```bash
import layoutparser as lp
model = lp.PaddleDetectionLayoutModel(model_path="inference/ppyolov2_r50vd_dcn_365e_coco", threshold=0.5,label_map={0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"},enforce_cpu=True,enable_mkldnn=True)
```
***
更多PaddleDetection训练教程，请参考：[PaddleDetection训练](https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.1/docs/tutorials/GETTING_STARTED_cn.md)
***
test1/paddlestructure.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/predict_system.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/setup.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/README.md 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/README_ch.md 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/__init__.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/eval_table.py 0 → 100755
浏览文件 @ 253b8453
此差异已折叠。
test1/table/matcher.py 0 → 100755
浏览文件 @ 253b8453
此差异已折叠。
test1/table/predict_structure.py 0 → 100755
浏览文件 @ 253b8453
此差异已折叠。
test1/table/predict_table.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/table_metric/__init__.py 0 → 100755
浏览文件 @ 253b8453
此差异已折叠。
test1/table/table_metric/parallel.py 0 → 100755
浏览文件 @ 253b8453
此差异已折叠。
test1/table/table_metric/table_metric.py 0 → 100755
浏览文件 @ 253b8453
此差异已折叠。
test1/table/tablepyxl/__init__.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/tablepyxl/style.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/table/tablepyxl/tablepyxl.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
test1/utility.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
tools/eval.py
浏览文件 @ 253b8453
此差异已折叠。
tools/export_model.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer/benchmark_utils.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
tools/infer/predict_cls.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer/predict_det.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer/predict_rec.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer/predict_system.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer/utility.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer_cls.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer_det.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer_e2e.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer_rec.py
浏览文件 @ 253b8453
此差异已折叠。
tools/infer_table.py 0 → 100644
浏览文件 @ 253b8453
此差异已折叠。
tools/program.py
浏览文件 @ 253b8453
此差异已折叠。
tools/train.py
浏览文件 @ 253b8453
此差异已折叠。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或