diff --git a/deploy/pdserving/README.md b/deploy/pdserving/README.md
index 88426ba9c508a4020af0a6203010d683cb73eba9..cb5e4bfa014a699daa492523918d2ed42fc6cd28 100644
--- a/deploy/pdserving/README.md
+++ b/deploy/pdserving/README.md
@@ -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:
+ 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]
+ ```
+
+
## FAQ
**Q1**: No result return after sending the request.
diff --git a/deploy/pdserving/README_CN.md b/deploy/pdserving/README_CN.md
index 3e3f1bde0e824fe6133a1c169b9b03e614904c26..1e53cb639ce903b7a42840f2e769e63f6894e2ce 100644
--- a/deploy/pdserving/README_CN.md
+++ b/deploy/pdserving/README_CN.md
@@ -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：
+ 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]
+ ```
+
+
## FAQ
diff --git a/deploy/pdserving/config.yml b/deploy/pdserving/config.yml
index aef735dbfab5b314f9209a7cc91e7fd5b6fc615c..2aae922dfa12f46d1c0ebd352e8d3a7077065cf8 100644
--- a/deploy/pdserving/config.yml
+++ b/deploy/pdserving/config.yml
@@ -1,32 +1,32 @@
#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
diff --git a/deploy/pdserving/ocr_reader.py b/deploy/pdserving/ocr_reader.py
index 95110706af13662de11ef0f668558d0dd3abcf52..3f219784fca79715d09ae9353a32d95e2e427cb6 100644
--- a/deploy/pdserving/ocr_reader.py
+++ b/deploy/pdserving/ocr_reader.py
@@ -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)
diff --git a/deploy/pdserving/pipeline_http_client.py b/deploy/pdserving/pipeline_http_client.py
index 88c4a81ea8bbed80d37b5fbfea6bf01b38f9613a..0befe2f6144d18e24fb3f72ed1d919fd8cd7d5a4 100644
--- a/deploy/pdserving/pipeline_http_client.py
+++ b/deploy/pdserving/pipeline_http_client.py
@@ -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)))
diff --git a/deploy/pdserving/pipeline_rpc_client.py b/deploy/pdserving/pipeline_rpc_client.py
index 7471f7ed6c1254d550bcf2c19f6ee7c610a2e20e..79f898faf37f946cdbf4a87d4d62c8b1f9d5c93b 100644
--- a/deploy/pdserving/pipeline_rpc_client.py
+++ b/deploy/pdserving/pipeline_rpc_client.py
@@ -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)
diff --git a/deploy/pdserving/web_service.py b/deploy/pdserving/web_service.py
index b47ef65d09dd7aad0e4d00ca852a5c32161ad45b..3f77f03537bce6df980abd8af83e7ed772e44d98 100644
--- a/deploy/pdserving/web_service.py
+++ b/deploy/pdserving/web_service.py
@@ -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, ""