Merge remote-tracking branch 'upstream/develop' into develop

deploy/cpp_infer/include/preprocess_op.h

@@ -62,4 +62,4 @@ public:
                    const std::vector<int> &rec_image_shape = {3, 48, 192});
 };
 
-} // namespace PaddleOCR
+} // namespace PaddleOCR
\ No newline at end of file

deploy/cpp_infer/src/preprocess_op.cpp

@@ -81,7 +81,7 @@ void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
   else if (resize_h / 32 < 1 + 1e-5)
     resize_h = 32;
   else
-    resize_h = (resize_h / 32 - 1) * 32;
+    resize_h = resize_h / 32 * 32;
 
   if (resize_w % 32 == 0)
     resize_w = resize_w;
@@ -96,13 +96,17 @@ void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
   ratio_w = float(resize_w) / float(w);
 }
 
-void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
+void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &pad_resize_img,
+                        float max_wh_ratio,
                         const std::vector<int> &rec_image_shape) {
   int imgC, imgH, imgW;
   imgC = rec_image_shape[0];
   imgH = rec_image_shape[1];
   imgW = rec_image_shape[2];
 
+  float wh_ratio = 1.0 * imgW / imgH;
+  wh_ratio = std::max(max_wh_ratio, wh_ratio);
+
   imgW = int(32 * wh_ratio);
 
   float ratio = float(img.cols) / float(img.rows);
@@ -112,8 +116,12 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
   else
     resize_w = int(ceilf(imgH * ratio));
 
+  cv::Mat resize_img;
   cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
              cv::INTER_LINEAR);
+  cv::copyMakeBorder(resize_img, pad_resize_img, 0, 0, 0,
+                     int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
+                     {127, 127, 127});
 }
 
 void ClsResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,

deploy/lite/readme.md

@@ -24,15 +24,7 @@ Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理
 ### 1.2 准备预测库
 
 预测库有两种获取方式：
-- 1. 直接下载，预测库下载链接如下：
-      |平台|预测库下载链接|
-      |-|-|
-      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
-      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
-
-      注：1. 上述预测库为PaddleLite 2.6.3分支编译得到，有关PaddleLite 2.6.3 详细信息可参考[链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.6.3)。
-
-- 2. [推荐]编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
+- 1. [推荐]编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
@@ -45,6 +37,9 @@ git checkout release/v2.7
 更多编译命令
 介绍请参考[链接](https://paddle-lite.readthedocs.io/zh/latest/user_guides/Compile/Android.html#id2)。
 
+- 2. 直接下载预测库，下载[链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.7.1)。
+
+
 直接下载预测库并解压后，可以得到`inference_lite_lib.android.armv8/`文件夹，通过编译Paddle-Lite得到的预测库位于
 `Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/`文件夹下。
 预测库的文件目录如下：
@@ -88,11 +83,11 @@ Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括
 |-|-|-|-|-|-|-|
 |V1.1|超轻量中文OCR 移动端模型|8.1M|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_det_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_cls_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_rec_opt.nb)|v2.7|
 |【slim】V1.1|超轻量中文OCR 移动端模型|3.5M|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_det_prune_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_cls_quant_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_rec_quant_opt.nb)|v2.7|
-|V1.0|轻量级中文OCR 移动端模型|8.6M|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.0_det_opt.nb)|---|[下载地址](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.0_rec_opt.nb)|v2.7|
+
 
 注意：V1.1 3.0M 轻量模型是使用PaddleSlim优化后的，需要配合Paddle-Lite最新预测库使用。
 
-如果直接使用上述表格中的模型进行部署，可略过下述步骤，直接阅读 [2.2节](#2.2与手机联调)。
+如果直接使用上述表格中的模型进行部署没有问题，可略过下述步骤，直接阅读 [2.2节](#2.2与手机联调)。
 
 如果要部署的模型不在上述表格中，则需要按照如下步骤获得优化后的模型。
 
@@ -184,7 +179,7 @@ wget  https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn_infer.tar && tar
  ```
  git clone https://github.com/PaddlePaddle/PaddleOCR.git
  cd PaddleOCR/deploy/lite/
- # 运行prepare.sh，准备预测库文件、测试图像和使用的字典文件，并放置在预测库中的demo/cxx/ocr文件夹下
+ # 按照如下命令运行prepare.sh，将预测库文件、测试图像和使用的字典文件到预测库中的demo/cxx/ocr文件夹下
  sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
 
  # 进入OCR demo的工作目录
@@ -255,6 +250,7 @@ use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1
  adb shell
  cd /data/local/tmp/debug
  export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
+ # ./ocr_db_crnn  检测模型文件 方向分类器模型文件  识别模型文件  测试图像路径  字典文件路径
  ./ocr_db_crnn ch_ppocr_mobile_v1.1_det_prune_opt.nb  ch_ppocr_mobile_v1.1_rec_quant_opt.nb  ch_ppocr_mobile_cls_quant_opt.nb  ./11.jpg  ppocr_keys_v1.txt
  ```
 

deploy/lite/readme_en.md

@@ -22,15 +22,7 @@ deployment solutions for end-side deployment issues.
 
 ## 3. Prepare prebuild library for android and ios
 
-### 3.1 Download prebuild library
-|Platform|Prebuild library Download Link|
-|-|-|
-|Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
-|IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.6.3/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
-
-note: The above pre-build inference library is compiled from the PaddleLite `release/v2.7` branch. For more information about PaddleLite 2.6.3, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.6.3).
-
-### 3.2 Compile prebuild library (Recommended)
+### 3.1 Compile prebuild library (Recommended)
 ```
 git clone https://github.com/PaddlePaddle/Paddle-Lite.git
 cd Paddle-Lite
@@ -66,6 +58,11 @@ inference_lite_lib.android.armv8/
 |   `-- java  
 ```
 
+### 3.2 Download prebuild library
+
+PaddleLite also provides a compiled [prediction library](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.7.1), developers can try on their own.
+
+
 
 ## 4. Inference Model Optimization
 
@@ -80,7 +77,6 @@ You can directly download the optimized model.
 | - | - | - | - | - | - | - |
 | V1.1 | extra-lightweight chinese OCR optimized model | 8.1M | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_det_opt.nb) | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_cls_opt.nb) | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_rec_opt.nb) | develop |
 | [slim] V1.1 | extra-lightweight chinese OCR optimized model | 3.5M | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_det_prune_opt.nb) | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_cls_quant_opt.nb) | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.1_rec_quant_opt.nb) | develop |
-| V1.0 | lightweight Chinese OCR optimized model | 8.6M | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.0_det_opt.nb) | - | [Download](https://paddleocr.bj.bcebos.com/20-09-22/mobile/lite/ch_ppocr_mobile_v1.0_rec_opt.nb) | develop |
 
 If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
 

doc/doc_en/algorithm_overview_en.md

@@ -33,7 +33,7 @@ On Total-Text dataset, the text detection result is as follows:
 
 **Note：** Additional data, like icdar2013, icdar2017, COCO-Text, ArT, was added to the model training of SAST. Download English public dataset in organized format used by PaddleOCR from [Baidu Drive](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (download code: 2bpi).
 
-For the training guide and use of PaddleOCR text detection algorithms, please refer to the document [Text detection model training/evaluation/prediction](./doc/doc_en/detection_en.md)
+For the training guide and use of PaddleOCR text detection algorithms, please refer to the document [Text detection model training/evaluation/prediction](./detection_en.md)
 
 <a name="TEXTRECOGNITIONALGORITHM"></a>
 ### 2. Text Recognition Algorithm
@@ -63,4 +63,4 @@ Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation r
 
 The average accuracy of the two-stage training in the original paper is 89.74%, and that of one stage training in paddleocr is 88.33%. Both pre-trained weights can be downloaded [here](https://paddleocr.bj.bcebos.com/SRN/rec_r50fpn_vd_none_srn.tar).
 
-Please refer to the document for training guide and use of PaddleOCR text recognition algorithms [Text recognition model training/evaluation/prediction](./doc/doc_en/recognition_en.md)
+Please refer to the document for training guide and use of PaddleOCR text recognition algorithms [Text recognition model training/evaluation/prediction](./recognition_en.md)

ppocr/modeling/backbones/det_mobilenet_v3.py

@@ -32,6 +32,7 @@ class MobileNetV3():
         """
         self.scale = params['scale']
         model_name = params['model_name']
+        self.model_name = model_name
         self.inplanes = 16
         if model_name == "large":
             self.cfg = [
@@ -80,7 +81,7 @@ class MobileNetV3():
             "supported scale are {} but input scale is {}".format(supported_scale, self.scale)
 
         self.disable_se = params.get('disable_se', False)
-        
+
     def __call__(self, input):
         scale = self.scale
         inplanes = self.inplanes
@@ -102,7 +103,8 @@ class MobileNetV3():
         inplanes = self.make_divisible(inplanes * scale)
         outs = []
         for layer_cfg in cfg:
-            if layer_cfg[5] == 2 and i > 2:
+            start_idx = 2 if self.model_name == 'large' else 0
+            if layer_cfg[5] == 2 and i > start_idx:
                 outs.append(conv)
             conv = self.residual_unit(
                 input=conv,

tools/infer/predict_det.py

@@ -200,4 +200,4 @@ if __name__ == "__main__":
         logger.info("The visualized img saved in {}".format(
             os.path.join(draw_img_save, "det_res_%s" % img_name_pure)))
     if count > 1:
-        logger.info("Avg Time:", total_time / (count - 1))
+        logger.info("Avg Time: {}".format(total_time / (count - 1)))