Skip to content

P
PaddleOCR

PaddlePaddle / PaddleOCR
10 个月 前同步成功

通知 1513
Star 32962
Fork 6643
P
PaddleOCR

前往新版Gitcode，体验更适合开发者的 AI 搜索 >>

未验证 提交 426e3e44 编写于 作者: M MissPenguin 提交者： GitHub
浏览文件
操作

Merge pull request #2771 from tink2123/pdserving 

update pdserving
上级 bcf59095 4135eb0c
隐藏空白更改
内联 并排
Showing with 235 addition and 106 deletion
deploy/pdserving/README.md
浏览文件 @ 426e3e44
......@@ -30,6 +30,8 @@ The introduction and tutorial of Paddle Serving service deployment framework ref
PaddleOCR operating environment and Paddle Serving operating environment are needed.
1. Please prepare PaddleOCR operating environment reference [link](../../doc/doc_ch/installation.md).
Download the corresponding paddle whl package according to the environment, it is recommended to install version 2.0.1.
2. The steps of PaddleServing operating environment prepare are as follows:
......@@ -45,23 +47,17 @@ PaddleOCR operating environment and Paddle Serving operating environment are nee
```
3. Install the client to send requests to the service
```
pip3 install paddle-serving-client==0.5.0 # for CPU
In [download link](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md) find the client installation package corresponding to the python version.
The python3.7 version is recommended here:
pip3 install paddle-serving-client-gpu==0.5.0 # for GPU
```
wget https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp37-none-any.whl
pip3 install paddle_serving_client-0.0.0-cp37-none-any.whl
```
4. Install serving-app
```
pip3 install paddle-serving-app==0.3.0
# fix local_predict to support load dynamic model
# find the install directoory of paddle_serving_app
vim /usr/local/lib/python3.7/site-packages/paddle_serving_app/local_predict.py
# replace line 85 of local_predict.py config = AnalysisConfig(model_path) with:
if os.path.exists(os.path.join(model_path, "__params__")):
config = AnalysisConfig(os.path.join(model_path, "__model__"), os.path.join(model_path, "__params__"))
else:
config = AnalysisConfig(model_path)
pip3 install paddle-serving-app==0.3.1
```
**note:** If you want to install the latest version of PaddleServing, refer to [link](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md).
......@@ -74,38 +70,38 @@ When using PaddleServing for service deployment, you need to convert the saved i
Firstly, download the [inference model](https://github.com/PaddlePaddle/PaddleOCR#pp-ocr-20-series-model-listupdate-on-dec-15) of PPOCR
```
# Download and unzip the OCR text detection model
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_det_infer.tar
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_det_infer.tar
# Download and unzip the OCR text recognition model
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar
```
Then, you can use installed paddle_serving_client tool to convert inference model to server model.
Then, you can use installed paddle_serving_client tool to convert inference model to mobile model.
```
# Detection model conversion
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_det_infer/ \
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_mobile_v2.0_det_infer/ \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--serving_server ./ppocr_det_server_2.0_serving/ \
--serving_client ./ppocr_det_server_2.0_client/
--serving_server ./ppocr_det_mobile_2.0_serving/ \
--serving_client ./ppocr_det_mobile_2.0_client/
# Recognition model conversion
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_rec_infer/ \
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_mobile_v2.0_rec_infer/ \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--serving_server ./ppocr_rec_server_2.0_serving/ \
--serving_client ./ppocr_rec_server_2.0_client/
--serving_server ./ppocr_rec_mobile_2.0_serving/ \
--serving_client ./ppocr_rec_mobile_2.0_client/
```
After the detection model is converted, there will be additional folders of `ppocr_det_server_2.0_serving` and `ppocr_det_server_2.0_client` in the current folder, with the following format:
After the detection model is converted, there will be additional folders of `ppocr_det_mobile_2.0_serving` and `ppocr_det_mobile_2.0_client` in the current folder, with the following format:
```
|- ppocr_det_server_2.0_serving/
|- ppocr_det_mobile_2.0_serving/
|- __model__
|- __params__
|- serving_server_conf.prototxt
|- serving_server_conf.stream.prototxt
|- ppocr_det_server_2.0_client
|- ppocr_det_mobile_2.0_client
|- serving_client_conf.prototxt
|- serving_client_conf.stream.prototxt
......@@ -147,6 +143,61 @@ The recognition model is the same.
After successfully running, the predicted result of the model will be printed in the cmd window. An example of the result is:
![](./imgs/results.png)
Adjust the number of concurrency in config.yml to get the largest QPS. Generally, the number of concurrent detection and recognition is 2:1
```
det:
concurrency: 8
...
rec:
concurrency: 4
...
```
Multiple service requests can be sent at the same time if necessary.
The predicted performance data will be automatically written into the `PipelineServingLogs/pipeline.tracer` file.
Tested on 200 real pictures, and limited the detection long side to 960. The average QPS on T4 GPU can reach around 23:
```
2021-05-13 03:42:36,895 ==================== TRACER ======================
2021-05-13 03:42:36,975 Op(rec):
2021-05-13 03:42:36,976 in[14.472382882882883 ms]
2021-05-13 03:42:36,976 prep[9.556855855855856 ms]
2021-05-13 03:42:36,976 midp[59.921905405405404 ms]
2021-05-13 03:42:36,976 postp[15.345945945945946 ms]
2021-05-13 03:42:36,976 out[1.9921216216216215 ms]
2021-05-13 03:42:36,976 idle[0.16254943864471572]
2021-05-13 03:42:36,976 Op(det):
2021-05-13 03:42:36,976 in[315.4468035714286 ms]
2021-05-13 03:42:36,976 prep[69.5980625 ms]
2021-05-13 03:42:36,976 midp[18.989535714285715 ms]
2021-05-13 03:42:36,976 postp[18.857803571428573 ms]
2021-05-13 03:42:36,977 out[3.1337544642857145 ms]
2021-05-13 03:42:36,977 idle[0.7477961159203756]
2021-05-13 03:42:36,977 DAGExecutor:
2021-05-13 03:42:36,977 Query count[224]
2021-05-13 03:42:36,977 QPS[22.4 q/s]
2021-05-13 03:42:36,977 Succ[0.9910714285714286]
2021-05-13 03:42:36,977 Error req[169, 170]
2021-05-13 03:42:36,977 Latency:
2021-05-13 03:42:36,977 ave[535.1678348214285 ms]
2021-05-13 03:42:36,977 .50[172.651 ms]
2021-05-13 03:42:36,977 .60[187.904 ms]
2021-05-13 03:42:36,977 .70[245.675 ms]
2021-05-13 03:42:36,977 .80[526.684 ms]
2021-05-13 03:42:36,977 .90[854.596 ms]
2021-05-13 03:42:36,977 .95[1722.728 ms]
2021-05-13 03:42:36,977 .99[3990.292 ms]
2021-05-13 03:42:36,978 Channel (server worker num[10]):
2021-05-13 03:42:36,978 chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
2021-05-13 03:42:36,979 chl1(In: ['det'], Out: ['rec']) size[6/0]
2021-05-13 03:42:36,979 chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
```
<a name="faq"></a>
## FAQ
**Q1**: No result return after sending the request.
......
deploy/pdserving/README_CN.md
浏览文件 @ 426e3e44
......@@ -29,7 +29,8 @@ PaddleOCR提供2种服务部署方式：
需要准备PaddleOCR的运行环境和Paddle Serving的运行环境。
- 准备PaddleOCR的运行环境参考[链接](../../doc/doc_ch/installation.md)
- 准备PaddleOCR的运行环境[链接](../../doc/doc_ch/installation.md)
根据环境下载对应的paddle whl包，推荐安装2.0.1版本
- 准备PaddleServing的运行环境，步骤如下
......@@ -45,25 +46,16 @@ PaddleOCR提供2种服务部署方式：
```
2. 安装client，用于向服务发送请求
```
pip3 install paddle-serving-client==0.5.0 # for CPU
在[下载链接](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md)中找到对应python版本的client安装包，这里推荐python3.7版本：
pip3 install paddle-serving-client-gpu==0.5.0 # for GPU
```
wget https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp37-none-any.whl
pip3 install paddle_serving_client-0.0.0-cp37-none-any.whl
```
3. 安装serving-app
```
pip3 install paddle-serving-app==0.3.0
```
**note:** 安装0.3.0版本的serving-app后，为了能加载动态图模型，需要修改serving_app的源码，具体为：
```
# 找到paddle_serving_app的安装目录，找到并编辑local_predict.py文件
vim /usr/local/lib/python3.7/site-packages/paddle_serving_app/local_predict.py
# 将local_predict.py 的第85行 config = AnalysisConfig(model_path) 替换为：
if os.path.exists(os.path.join(model_path, "__params__")):
config = AnalysisConfig(os.path.join(model_path, "__model__"), os.path.join(model_path, "__params__"))
else:
config = AnalysisConfig(model_path)
pip3 install paddle-serving-app==0.3.1
```
**Note:** 如果要安装最新版本的PaddleServing参考[链接](https://github.com/PaddlePaddle/Serving/blob/develop/doc/LATEST_PACKAGES.md)。
......@@ -76,38 +68,38 @@ PaddleOCR提供2种服务部署方式：
首先，下载PPOCR的[inference模型](https://github.com/PaddlePaddle/PaddleOCR#pp-ocr-20-series-model-listupdate-on-dec-15)
```
# 下载并解压 OCR 文本检测模型
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_det_infer.tar
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_det_infer.tar
# 下载并解压 OCR 文本识别模型
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar
wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar
```
接下来，用安装的paddle_serving_client把下载的inference模型转换成易于server部署的模型格式。
```
# 转换检测模型
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_det_infer/ \
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_mobile_v2.0_det_infer/ \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--serving_server ./ppocr_det_server_2.0_serving/ \
--serving_client ./ppocr_det_server_2.0_client/
--serving_server ./ppocr_det_mobile_2.0_serving/ \
--serving_client ./ppocr_det_mobile_2.0_client/
# 转换识别模型
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_rec_infer/ \
python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_mobile_v2.0_rec_infer/ \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--serving_server ./ppocr_rec_server_2.0_serving/ \
--serving_client ./ppocr_rec_server_2.0_client/
--serving_server ./ppocr_rec_mobile_2.0_serving/ \
--serving_client ./ppocr_rec_mobile_2.0_client/
```
检测模型转换完成后，会在当前文件夹多出`ppocr_det_server_2.0_serving` 和`ppocr_det_server_2.0_client`的文件夹，具备如下格式：
检测模型转换完成后，会在当前文件夹多出`ppocr_det_mobile_2.0_serving` 和`ppocr_det_mobile_2.0_client`的文件夹，具备如下格式：
```
|- ppocr_det_server_2.0_serving/
|- ppocr_det_mobile_2.0_serving/
|- __model__
|- __params__
|- serving_server_conf.prototxt
|- serving_server_conf.stream.prototxt
|- ppocr_det_server_2.0_client
|- ppocr_det_mobile_2.0_client
|- serving_client_conf.prototxt
|- serving_client_conf.stream.prototxt
......@@ -148,6 +140,60 @@ python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_server_v2.0_rec_in
成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为：
![](./imgs/results.png)
调整 config.yml 中的并发个数获得最大的QPS, 一般检测和识别的并发数为2：1
```
det:
#并发数，is_thread_op=True时，为线程并发；否则为进程并发
concurrency: 8
...
rec:
#并发数，is_thread_op=True时，为线程并发；否则为进程并发
concurrency: 4
...
```
有需要的话可以同时发送多个服务请求
预测性能数据会被自动写入 `PipelineServingLogs/pipeline.tracer` 文件中。
在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 均值可达到23左右：
```
2021-05-13 03:42:36,895 ==================== TRACER ======================
2021-05-13 03:42:36,975 Op(rec):
2021-05-13 03:42:36,976 in[14.472382882882883 ms]
2021-05-13 03:42:36,976 prep[9.556855855855856 ms]
2021-05-13 03:42:36,976 midp[59.921905405405404 ms]
2021-05-13 03:42:36,976 postp[15.345945945945946 ms]
2021-05-13 03:42:36,976 out[1.9921216216216215 ms]
2021-05-13 03:42:36,976 idle[0.16254943864471572]
2021-05-13 03:42:36,976 Op(det):
2021-05-13 03:42:36,976 in[315.4468035714286 ms]
2021-05-13 03:42:36,976 prep[69.5980625 ms]
2021-05-13 03:42:36,976 midp[18.989535714285715 ms]
2021-05-13 03:42:36,976 postp[18.857803571428573 ms]
2021-05-13 03:42:36,977 out[3.1337544642857145 ms]
2021-05-13 03:42:36,977 idle[0.7477961159203756]
2021-05-13 03:42:36,977 DAGExecutor:
2021-05-13 03:42:36,977 Query count[224]
2021-05-13 03:42:36,977 QPS[22.4 q/s]
2021-05-13 03:42:36,977 Succ[0.9910714285714286]
2021-05-13 03:42:36,977 Error req[169, 170]
2021-05-13 03:42:36,977 Latency:
2021-05-13 03:42:36,977 ave[535.1678348214285 ms]
2021-05-13 03:42:36,977 .50[172.651 ms]
2021-05-13 03:42:36,977 .60[187.904 ms]
2021-05-13 03:42:36,977 .70[245.675 ms]
2021-05-13 03:42:36,977 .80[526.684 ms]
2021-05-13 03:42:36,977 .90[854.596 ms]
2021-05-13 03:42:36,977 .95[1722.728 ms]
2021-05-13 03:42:36,977 .99[3990.292 ms]
2021-05-13 03:42:36,978 Channel (server worker num[10]):
2021-05-13 03:42:36,978 chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
2021-05-13 03:42:36,979 chl1(In: ['det'], Out: ['rec']) size[6/0]
2021-05-13 03:42:36,979 chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
```
<a name="FAQ"></a>
## FAQ
......
deploy/pdserving/config.yml
浏览文件 @ 426e3e44
#rpc端口, rpc_port和http_port不允许同时为空。当rpc_port为空且http_port不为空时，会自动将rpc_port设置为http_port+1
rpc_port: 18090
rpc_port: 18091
#http端口, rpc_port和http_port不允许同时为空。当rpc_port可用且http_port为空时，不自动生成http_port
http_port: 9999
http_port: 9998
#worker_num, 最大并发数。当build_dag_each_worker=True时, 框架会创建worker_num个进程，每个进程内构建grpcSever和DAG
##当build_dag_each_worker=False时，框架会设置主线程grpc线程池的max_workers=worker_num
worker_num: 20
worker_num: 10
#build_dag_each_worker, False，框架在进程内创建一条DAG；True，框架会每个进程内创建多个独立的DAG
build_dag_each_worker: false
build_dag_each_worker: False
dag:
#op资源类型, True, 为线程模型；False，为进程模型
is_thread_op: False
#重试次数
retry: 1
retry: 10
#使用性能分析, True，生成Timeline性能数据，对性能有一定影响；False为不使用
use_profile: False
use_profile: True
tracer:
interval_s: 10
op:
det:
#并发数，is_thread_op=True时，为线程并发；否则为进程并发
concurrency: 4
concurrency: 8
#当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
local_service_conf:
......@@ -34,18 +34,18 @@ op:
client_type: local_predictor
#det模型路径
model_config: /paddle/serving/models/det_serving_server/ #ocr_det_model
model_config: ./ppocr_det_mobile_2.0_serving
#Fetch结果列表，以client_config中fetch_var的alias_name为准
fetch_list: ["save_infer_model/scale_0.tmp_1"]
#计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
devices: "2"
devices: "0"
ir_optim: True
rec:
#并发数，is_thread_op=True时，为线程并发；否则为进程并发
concurrency: 1
concurrency: 4
#超时时间, 单位ms
timeout: -1
......@@ -60,12 +60,12 @@ op:
client_type: local_predictor
#rec模型路径
model_config: /paddle/serving/models/rec_serving_server/ #ocr_rec_model
model_config: ./ppocr_rec_mobile_2.0_serving
#Fetch结果列表，以client_config中fetch_var的alias_name为准
fetch_list: ["save_infer_model/scale_0.tmp_1"] #["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"]
fetch_list: ["save_infer_model/scale_0.tmp_1"]
#计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
devices: "2"
devices: "0"
ir_optim: True
deploy/pdserving/ocr_reader.py
浏览文件 @ 426e3e44
......@@ -21,7 +21,6 @@ import sys
import argparse
import string
from copy import deepcopy
import paddle
class DetResizeForTest(object):
......@@ -34,12 +33,12 @@ class DetResizeForTest(object):
elif 'limit_side_len' in kwargs:
self.limit_side_len = kwargs['limit_side_len']
self.limit_type = kwargs.get('limit_type', 'min')
elif 'resize_long' in kwargs:
self.resize_type = 2
self.resize_long = kwargs.get('resize_long', 960)
else:
elif 'resize_short' in kwargs:
self.limit_side_len = 736
self.limit_type = 'min'
else:
self.resize_type = 2
self.resize_long = kwargs.get('resize_long', 960)
def __call__(self, data):
img = deepcopy(data)
......@@ -227,8 +226,6 @@ class CTCLabelDecode(BaseRecLabelDecode):
super(CTCLabelDecode, self).__init__(config)
def __call__(self, preds, label=None, *args, **kwargs):
if isinstance(preds, paddle.Tensor):
preds = preds.numpy()
preds_idx = preds.argmax(axis=2)
preds_prob = preds.max(axis=2)
text = self.decode(preds_idx, preds_prob, is_remove_duplicate=True)
......
deploy/pdserving/pipeline_http_client.py
浏览文件 @ 426e3e44
......@@ -23,8 +23,8 @@ def cv2_to_base64(image):
return base64.b64encode(image).decode('utf8')
url = "http://127.0.0.1:9999/ocr/prediction"
test_img_dir = "../doc/imgs/"
url = "http://127.0.0.1:9998/ocr/prediction"
test_img_dir = "../../doc/imgs/"
for idx, img_file in enumerate(os.listdir(test_img_dir)):
with open(os.path.join(test_img_dir, img_file), 'rb') as file:
image_data1 = file.read()
......@@ -36,5 +36,5 @@ for idx, img_file in enumerate(os.listdir(test_img_dir)):
r = requests.post(url=url, data=json.dumps(data))
print(r.json())
test_img_dir = "../doc/imgs/"
test_img_dir = "../../doc/imgs/"
print("==> total number of test imgs: ", len(os.listdir(test_img_dir)))
deploy/pdserving/pipeline_rpc_client.py
浏览文件 @ 426e3e44
......@@ -23,7 +23,7 @@ import base64
import os
client = PipelineClient()
client.connect(['127.0.0.1:18090'])
client.connect(['127.0.0.1:18091'])
def cv2_to_base64(image):
......@@ -39,4 +39,3 @@ for img_file in os.listdir(test_img_dir):
for i in range(1):
ret = client.predict(feed_dict={"image": image}, fetch=["res"])
print(ret)
#print(ret)
deploy/pdserving/web_service.py
浏览文件 @ 426e3e44
......@@ -48,28 +48,24 @@ class DetOp(Op):
def preprocess(self, input_dicts, data_id, log_id):
(_, input_dict), = input_dicts.items()
data = base64.b64decode(input_dict["image"].encode('utf8'))
self.raw_im = data
data = np.fromstring(data, np.uint8)
# Note: class variables(self.var) can only be used in process op mode
im = cv2.imdecode(data, cv2.IMREAD_COLOR)
self.im = im
self.ori_h, self.ori_w, _ = im.shape
det_img = self.det_preprocess(self.im)
det_img = self.det_preprocess(im)
_, self.new_h, self.new_w = det_img.shape
print("det image shape", det_img.shape)
return {"x": det_img[np.newaxis, :].copy()}, False, None, ""
def postprocess(self, input_dicts, fetch_dict, log_id):
print("input_dicts: ", input_dicts)
det_out = fetch_dict["save_infer_model/scale_0.tmp_1"]
ratio_list = [
float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
]
dt_boxes_list = self.post_func(det_out, [ratio_list])
dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
out_dict = {"dt_boxes": dt_boxes, "image": self.im}
out_dict = {"dt_boxes": dt_boxes, "image": self.raw_im}
print("out dict", out_dict["dt_boxes"])
return out_dict, None, ""
......@@ -83,35 +79,75 @@ class RecOp(Op):
def preprocess(self, input_dicts, data_id, log_id):
(_, input_dict), = input_dicts.items()
im = input_dict["image"]
raw_im = input_dict["image"]
data = np.frombuffer(raw_im, np.uint8)
im = cv2.imdecode(data, cv2.IMREAD_COLOR)
dt_boxes = input_dict["dt_boxes"]
dt_boxes = self.sorted_boxes(dt_boxes)
feed_list = []
img_list = []
max_wh_ratio = 0
for i, dtbox in enumerate(dt_boxes):
boximg = self.get_rotate_crop_image(im, dt_boxes[i])
img_list.append(boximg)
h, w = boximg.shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
_, w, h = self.ocr_reader.resize_norm_img(img_list[0],
max_wh_ratio).shape
imgs = np.zeros((len(img_list), 3, w, h)).astype('float32')
for id, img in enumerate(img_list):
norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
imgs[id] = norm_img
print("rec image shape", imgs.shape)
feed = {"x": imgs.copy()}
return feed, False, None, ""
def postprocess(self, input_dicts, fetch_dict, log_id):
rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
res_lst = []
for res in rec_res:
res_lst.append(res[0])
res = {"res": str(res_lst)}
## Many mini-batchs, the type of feed_data is list.
max_batch_size = 6 # len(dt_boxes)
# If max_batch_size is 0, skipping predict stage
if max_batch_size == 0:
return {}, True, None, ""
boxes_size = len(dt_boxes)
batch_size = boxes_size // max_batch_size
rem = boxes_size % max_batch_size
for bt_idx in range(0, batch_size + 1):
imgs = None
boxes_num_in_one_batch = 0
if bt_idx == batch_size:
if rem == 0:
continue
else:
boxes_num_in_one_batch = rem
elif bt_idx < batch_size:
boxes_num_in_one_batch = max_batch_size
else:
_LOGGER.error("batch_size error, bt_idx={}, batch_size={}".
format(bt_idx, batch_size))
break
start = bt_idx * max_batch_size
end = start + boxes_num_in_one_batch
img_list = []
for box_idx in range(start, end):
boximg = self.get_rotate_crop_image(im, dt_boxes[box_idx])
img_list.append(boximg)
h, w = boximg.shape[0:2]
wh_ratio = w * 1.0 / h
max_wh_ratio = max(max_wh_ratio, wh_ratio)
_, w, h = self.ocr_reader.resize_norm_img(img_list[0],
max_wh_ratio).shape
imgs = np.zeros((boxes_num_in_one_batch, 3, w, h)).astype('float32')
for id, img in enumerate(img_list):
norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
imgs[id] = norm_img
feed = {"x": imgs.copy()}
feed_list.append(feed)
return feed_list, False, None, ""
def postprocess(self, input_dicts, fetch_data, log_id):
res_list = []
if isinstance(fetch_data, dict):
if len(fetch_data) > 0:
rec_batch_res = self.ocr_reader.postprocess(
fetch_data, with_score=True)
for res in rec_batch_res:
res_list.append(res[0])
elif isinstance(fetch_data, list):
for one_batch in fetch_data:
one_batch_res = self.ocr_reader.postprocess(
one_batch, with_score=True)
for res in one_batch_res:
res_list.append(res[0])
res = {"res": str(res_list)}
return res, None, ""
......
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