diff --git a/deploy/paddleserving/imgs/results_recog.png b/deploy/paddleserving/imgs/results_recog.png
deleted file mode 100644
index 37393d5d64e84de469d78dcc9fad88aa771f57f8..0000000000000000000000000000000000000000
Binary files a/deploy/paddleserving/imgs/results_recog.png and /dev/null differ
diff --git a/deploy/paddleserving/imgs/results_shitu.png b/deploy/paddleserving/imgs/results_shitu.png
new file mode 100644
index 0000000000000000000000000000000000000000..b8dc2480557aed6f56e8067324d450444ab37c73
Binary files /dev/null and b/deploy/paddleserving/imgs/results_shitu.png differ
diff --git a/deploy/paddleserving/imgs/start_server_recog.png b/deploy/paddleserving/imgs/start_server_recog.png
deleted file mode 100644
index d4344a1e6bdab7ccc4c3c31bc16d1e3186b9b806..0000000000000000000000000000000000000000
Binary files a/deploy/paddleserving/imgs/start_server_recog.png and /dev/null differ
diff --git a/deploy/paddleserving/imgs/start_server_shitu.png b/deploy/paddleserving/imgs/start_server_shitu.png
new file mode 100644
index 0000000000000000000000000000000000000000..32eb0cfe6757ed9f941dbf7aec7ffc96e6f45a2f
Binary files /dev/null and b/deploy/paddleserving/imgs/start_server_shitu.png differ
diff --git a/deploy/paddleserving/recognition/config.yml b/deploy/paddleserving/recognition/config.yml
index f67ee5521b5135b28f0b945cf800f477e18cc787..9e4a79ec47e88cab4a3f7d2d84d348167642c0d0 100644
--- a/deploy/paddleserving/recognition/config.yml
+++ b/deploy/paddleserving/recognition/config.yml
@@ -18,7 +18,7 @@ op:
local_service_conf:
#uci模型路径
- model_config: ../../models/product_ResNet50_vd_aliproduct_v1.0_serving
+ model_config: ../../models/general_PPLCNet_x2_5_lite_v1.0_serving
#计算硬件类型: 空缺时由devices决定(CPU/GPU)，0=cpu, 1=gpu, 2=tensorRT, 3=arm cpu, 4=kunlun xpu
device_type: 1
@@ -40,4 +40,4 @@ op:
devices: '0'
fetch_list:
- save_infer_model/scale_0.tmp_1
- model_config: ../../models/ppyolov2_r50vd_dcn_mainbody_v1.0_serving/
\ No newline at end of file
+ model_config: ../../models/picodet_PPLCNet_x2_5_mainbody_lite_v1.0_serving/
\ No newline at end of file
diff --git a/deploy/paddleserving/recognition/pipeline_http_client.py b/deploy/paddleserving/recognition/pipeline_http_client.py
index aa0cb5429b7a5ef06567a1655d6f5d89f489ecea..7e92fc6c03f733b1841078e77fefb7fa62318540 100644
--- a/deploy/paddleserving/recognition/pipeline_http_client.py
+++ b/deploy/paddleserving/recognition/pipeline_http_client.py
@@ -3,15 +3,17 @@ import json
import base64
import os
-imgpath = "daoxiangcunjinzhubing_6.jpg"
+imgpath = "../../drink_dataset_v1.0/test_images/001.jpeg"
+
def cv2_to_base64(image):
return base64.b64encode(image).decode('utf8')
+
if __name__ == "__main__":
url = "http://127.0.0.1:18081/recognition/prediction"
- with open(os.path.join(".", imgpath), 'rb') as file:
+ with open(os.path.join(".", imgpath), 'rb') as file:
image_data1 = file.read()
image = cv2_to_base64(image_data1)
data = {"key": ["image"], "value": [image]}
diff --git a/deploy/paddleserving/recognition/recognition_web_service.py b/deploy/paddleserving/recognition/recognition_web_service.py
index 88daf96e6ba2ea9b5bf030f2a2bd83c6d645bd5e..3425e23fba5c2e27ae9299c314e0ed7d6a0ff121 100644
--- a/deploy/paddleserving/recognition/recognition_web_service.py
+++ b/deploy/paddleserving/recognition/recognition_web_service.py
@@ -23,6 +23,7 @@ import faiss
import pickle
import json
+
class DetOp(Op):
def init_op(self):
self.img_preprocess = Sequential([
@@ -62,50 +63,57 @@ class DetOp(Op):
im_scale_y, im_scale_x = self.generate_scale(raw_im)
im = self.img_preprocess(raw_im)
-
+
imgs.append({
- "image": im[np.newaxis, :],
- "im_shape": np.array(list(im.shape[1:])).reshape(-1)[np.newaxis,:],
- "scale_factor": np.array([im_scale_y, im_scale_x]).astype('float32'),
+ "image": im[np.newaxis, :],
+ "im_shape":
+ np.array(list(im.shape[1:])).reshape(-1)[np.newaxis, :],
+ "scale_factor":
+ np.array([im_scale_y, im_scale_x]).astype('float32'),
})
self.raw_img = raw_imgs
feed_dict = {
- "image": np.concatenate([x["image"] for x in imgs], axis=0),
- "im_shape": np.concatenate([x["im_shape"] for x in imgs], axis=0),
- "scale_factor": np.concatenate([x["scale_factor"] for x in imgs], axis=0)
+ "image": np.concatenate(
+ [x["image"] for x in imgs], axis=0),
+ "im_shape": np.concatenate(
+ [x["im_shape"] for x in imgs], axis=0),
+ "scale_factor": np.concatenate(
+ [x["scale_factor"] for x in imgs], axis=0)
}
- return feed_dict, False, None, ""
+ return feed_dict, False, None, ""
def postprocess(self, input_dicts, fetch_dict, log_id):
boxes = self.img_postprocess(fetch_dict, visualize=False)
- boxes.sort(key = lambda x: x["score"], reverse = True)
- boxes = filter(lambda x: x["score"] >= self.threshold, boxes[:self.max_det_results])
+ boxes.sort(key=lambda x: x["score"], reverse=True)
+ boxes = filter(lambda x: x["score"] >= self.threshold,
+ boxes[:self.max_det_results])
boxes = list(boxes)
for i in range(len(boxes)):
boxes[i]["bbox"][2] += boxes[i]["bbox"][0] - 1
boxes[i]["bbox"][3] += boxes[i]["bbox"][1] - 1
result = json.dumps(boxes)
res_dict = {"bbox_result": result, "image": self.raw_img}
- return res_dict, None, ""
+ return res_dict, None, ""
+
class RecOp(Op):
def init_op(self):
self.seq = Sequential([
- BGR2RGB(), Resize((224, 224)),
- Div(255), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225],
- False), Transpose((2, 0, 1))
+ BGR2RGB(), Resize((224, 224)), Div(255),
+ Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225],
+ False), Transpose((2, 0, 1))
])
- index_dir = "../../recognition_demo_data_v1.1/gallery_product/index"
+ index_dir = "../../drink_dataset_v1.0/index"
assert os.path.exists(os.path.join(
index_dir, "vector.index")), "vector.index not found ..."
assert os.path.exists(os.path.join(
index_dir, "id_map.pkl")), "id_map.pkl not found ... "
-
+
self.searcher = faiss.read_index(
os.path.join(index_dir, "vector.index"))
-
+
with open(os.path.join(index_dir, "id_map.pkl"), "rb") as fd:
self.id_map = pickle.load(fd)
@@ -121,24 +129,25 @@ class RecOp(Op):
origin_img = cv2.imdecode(data, cv2.IMREAD_COLOR)
dt_boxes = input_dict["bbox_result"]
boxes = json.loads(dt_boxes)
- boxes.append({"category_id": 0,
- "score": 1.0,
- "bbox": [0, 0, origin_img.shape[1], origin_img.shape[0]]
- })
+ boxes.append({
+ "category_id": 0,
+ "score": 1.0,
+ "bbox": [0, 0, origin_img.shape[1], origin_img.shape[0]]
+ })
self.det_boxes = boxes
#construct batch images for rec
imgs = []
for box in boxes:
box = [int(x) for x in box["bbox"]]
- im = origin_img[box[1]: box[3], box[0]: box[2]].copy()
+ im = origin_img[box[1]:box[3], box[0]:box[2]].copy()
img = self.seq(im)
imgs.append(img[np.newaxis, :].copy())
input_imgs = np.concatenate(imgs, axis=0)
- return {"x": input_imgs}, False, None, ""
+ return {"x": input_imgs}, False, None, ""
- def nms_to_rec_results(self, results, thresh = 0.1):
+ def nms_to_rec_results(self, results, thresh=0.1):
filtered_results = []
x1 = np.array([r["bbox"][0] for r in results]).astype("float32")
y1 = np.array([r["bbox"][1] for r in results]).astype("float32")
@@ -172,7 +181,7 @@ class RecOp(Op):
np.sum(np.square(batch_features), axis=1, keepdims=True))
batch_features = np.divide(batch_features, feas_norm)
- scores, docs = self.searcher.search(batch_features, self.return_k)
+ scores, docs = self.searcher.search(batch_features, self.return_k)
results = []
for i in range(scores.shape[0]):
@@ -182,17 +191,19 @@ class RecOp(Op):
pred["rec_docs"] = self.id_map[docs[i][0]].split()[1]
pred["rec_scores"] = scores[i][0]
results.append(pred)
-
+
#do nms
results = self.nms_to_rec_results(results, self.rec_nms_thresold)
return {"result": str(results)}, None, ""
+
class RecognitionService(WebService):
def get_pipeline_response(self, read_op):
det_op = DetOp(name="det", input_ops=[read_op])
rec_op = RecOp(name="rec", input_ops=[det_op])
return rec_op
+
product_recog_service = RecognitionService(name="recognition")
product_recog_service.prepare_pipeline_config("config.yml")
product_recog_service.run_service()
diff --git a/docs/zh_CN/inference_deployment/paddle_serving_deploy.md b/docs/zh_CN/inference_deployment/paddle_serving_deploy.md
index 94f8906c4d7ce84f25731363824d12c9ac9283fb..74bbc20a7896981afecb577b29156eda6c027b55 100644
--- a/docs/zh_CN/inference_deployment/paddle_serving_deploy.md
+++ b/docs/zh_CN/inference_deployment/paddle_serving_deploy.md
@@ -1,11 +1,17 @@
# 模型服务化部署
+- [简介](#简介)
+- [Serving安装](#Serving安装)
+- [图像分类服务部署](#图像分类服务部署)
+- [图像识别服务部署](#图像识别服务部署)
+- [FAQ](#FAQ)
+
## 1. 简介
[Paddle Serving](https://github.com/PaddlePaddle/Serving) 旨在帮助深度学习开发者轻松部署在线预测服务，支持一键部署工业级的服务能力、客户端和服务端之间高并发和高效通信、并支持多种编程语言开发客户端。
该部分以 HTTP 预测服务部署为例，介绍怎样在 PaddleClas 中使用 PaddleServing 部署模型服务。
-
+
## 2. Serving安装
Serving 官网推荐使用 docker 安装并部署 Serving 环境。首先需要拉取 docker 环境并创建基于 Serving 的 docker。
@@ -22,6 +28,7 @@ nvidia-docker exec -it test bash
pip install paddlepaddle-gpu
pip install paddle-serving-client
pip install paddle-serving-server-gpu
+pip install paddle-serving-app
```
* 如果安装速度太慢，可以通过 `-i https://pypi.tuna.tsinghua.edu.cn/simple` 更换源，加速安装过程。
@@ -31,35 +38,186 @@ pip install paddle-serving-server-gpu
```shell
pip install paddle-serving-server
```
-
-## 3. 导出模型
-
-使用 `tools/export_serving_model.py` 脚本导出 Serving 模型，以 `ResNet50_vd` 为例，使用方法如下。
-
+
+## 3. 图像分类服务部署
+### 3.1 模型转换
+使用PaddleServing做服务化部署时，需要将保存的inference模型转换为Serving模型。下面以经典的ResNet50_vd模型为例，介绍如何部署图像分类服务。
+- 进入工作目录：
```shell
-python tools/export_serving_model.py -m ResNet50_vd -p ./pretrained/ResNet50_vd_pretrained/ -o serving
+cd deploy/paddleserving
+```
+- 下载ResNet50_vd的inference模型：
+```shell
+# 下载并解压ResNet50_vd模型
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/ResNet50_vd_infer.tar && tar xf ResNet50_vd_infer.tar
+```
+- 用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ResNet50_vd模型
+python3 -m paddle_serving_client.convert --dirname ./ResNet50_vd_infer/ \
+ --model_filename inference.pdmodel \
+ --params_filename inference.pdiparams \
+ --serving_server ./ResNet50_vd_serving/ \
+ --serving_client ./ResNet50_vd_client/
+```
+ResNet50_vd推理模型转换完成后，会在当前文件夹多出`ResNet50_vd_serving` 和`ResNet50_vd_client`的文件夹，具备如下格式：
```
+|- ResNet50_vd_client/
+ |- __model__
+ |- __params__
+ |- serving_server_conf.prototxt
+ |- serving_server_conf.stream.prototxt
+|- ResNet50_vd_client
+ |- serving_client_conf.prototxt
+ |- serving_client_conf.stream.prototxt
+```
+得到模型文件之后，需要修改serving_server_conf.prototxt中的alias名字： 将`feed_var`中的`alias_name`改为`image`, 将`fetch_var`中的`alias_name`改为`prediction`
-最终在 serving 文件夹下会生成 `ppcls_client_conf` 与 `ppcls_model` 两个文件夹，分别存储了 client 配置、模型参数与结构文件。
+**备注**: Serving为了兼容不同模型的部署，提供了输入输出重命名的功能。这样，不同的模型在推理部署时，只需要修改配置文件的alias_name即可，无需修改代码即可完成推理部署。
+修改后的serving_server_conf.prototxt如下所示:
+```
+feed_var {
+ name: "inputs"
+ alias_name: "image"
+ is_lod_tensor: false
+ feed_type: 1
+ shape: 3
+ shape: 224
+ shape: 224
+}
+fetch_var {
+ name: "save_infer_model/scale_0.tmp_1"
+ alias_name: "prediction"
+ is_lod_tensor: true
+ fetch_type: 1
+ shape: -1
+}
+```
+### 3.2 服务部署和请求
+paddleserving目录包含了启动pipeline服务和发送预测请求的代码，包括：
+```shell
+__init__.py
+config.yml # 启动服务的配置文件
+pipeline_http_client.py # http方式发送pipeline预测请求的脚本
+pipeline_rpc_client.py # rpc方式发送pipeline预测请求的脚本
+classification_web_service.py # 启动pipeline服务端的脚本
+```
+- 启动服务：
+```shell
+# 启动服务，运行日志保存在log.txt
+python3 classification_web_service.py &>log.txt &
+```
+成功启动服务后，log.txt中会打印类似如下日志
+
-## 4. 服务部署与请求
+- 发送请求：
+```shell
+# 发送服务请求
+python3 pipeline_http_client.py
+```
+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：
+
+
+
+## 4.图像识别服务部署
+使用PaddleServing做服务化部署时，需要将保存的inference模型转换为Serving模型。 下面以PP-ShiTu中的超轻量图像识别模型为例，介绍图像识别服务的部署。
+## 4.1 模型转换
+- 下载通用检测inference模型和通用识别inference模型
+```
+cd deploy
+# 下载并解压通用识别模型
+wget -P models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/inference/general_PPLCNet_x2_5_lite_v1.0_infer.tar
+cd models
+tar -xf general_PPLCNet_x2_5_lite_v1.0_infer.tar
+# 下载并解压通用检测模型
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/inference/picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer.tar
+tar -xf picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer.tar
+```
+- 转换识别inference模型为Serving模型：
+```
+# 转换识别模型
+python3 -m paddle_serving_client.convert --dirname ./general_PPLCNet_x2_5_lite_v1.0_infer/ \
+ --model_filename inference.pdmodel \
+ --params_filename inference.pdiparams \
+ --serving_server ./general_PPLCNet_x2_5_lite_v1.0_serving/ \
+ --serving_client ./general_PPLCNet_x2_5_lite_v1.0_client/
+```
+识别推理模型转换完成后，会在当前文件夹多出`general_PPLCNet_x2_5_lite_v1.0_serving/` 和`general_PPLCNet_x2_5_lite_v1.0_serving/`的文件夹。修改`general_PPLCNet_x2_5_lite_v1.0_serving/`目录下的serving_server_conf.prototxt中的alias名字： 将`fetch_var`中的`alias_name`改为`features`。
+修改后的serving_server_conf.prototxt内容如下：
+```
+feed_var {
+ name: "x"
+ alias_name: "x"
+ is_lod_tensor: false
+ feed_type: 1
+ shape: 3
+ shape: 224
+ shape: 224
+}
+fetch_var {
+ name: "save_infer_model/scale_0.tmp_1"
+ alias_name: "features"
+ is_lod_tensor: true
+ fetch_type: 1
+ shape: -1
+}
+```
+- 转换通用检测inference模型为Serving模型：
+```
+# 转换通用检测模型
+python3 -m paddle_serving_client.convert --dirname ./picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer/ \
+ --model_filename inference.pdmodel \
+ --params_filename inference.pdiparams \
+ --serving_server ./picodet_PPLCNet_x2_5_mainbody_lite_v1.0_serving/ \
+ --serving_client ./picodet_PPLCNet_x2_5_mainbody_lite_v1.0_client/
+```
+检测inference模型转换完成后，会在当前文件夹多出`picodet_PPLCNet_x2_5_mainbody_lite_v1.0_serving/` 和`picodet_PPLCNet_x2_5_mainbody_lite_v1.0_client/`的文件夹。
-* 使用下面的方式启动 Serving 服务。
+**注意:** 此处不需要修改`picodet_PPLCNet_x2_5_mainbody_lite_v1.0_serving/`目录下的serving_server_conf.prototxt中的alias名字。
+- 下载并解压已经构建后的检索库index
+```
+cd ../
+wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/drink_dataset_v1.0.tar && tar -xf drink_dataset_v1.0.tar
+```
+## 4.2 服务部署和请求
+**注意:** 识别服务涉及到多个模型，出于性能考虑采用PipeLine部署方式。Pipeline部署方式当前不支持windows平台。
+- 进入到工作目录
```shell
-python tools/serving/image_service_gpu.py serving/ppcls_model workdir 9292
+cd ./deploy/paddleserving/recognition
```
+paddleserving目录包含启动pipeline服务和发送预测请求的代码，包括：
+```
+__init__.py
+config.yml # 启动服务的配置文件
+pipeline_http_client.py # http方式发送pipeline预测请求的脚本
+pipeline_rpc_client.py # rpc方式发送pipeline预测请求的脚本
+recognition_web_service.py # 启动pipeline服务端的脚本
+```
+- 启动服务：
+```
+# 启动服务，运行日志保存在log.txt
+python3 recognition_web_service.py &>log.txt &
+```
+成功启动服务后，log.txt中会打印类似如下日志
+
-其中 `serving/ppcls_model` 为刚才保存的 Serving 模型地址，`workdir` 为工作目录，`9292` 为服务的端口号。
-
+- 发送请求：
+```
+python3 pipeline_http_client.py
+```
+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：
+
-* 使用下面的脚本向 Serving 服务发送识别请求，并返回结果。
+
+## 5.FAQ
+**Q1**： 发送请求后没有结果返回或者提示输出解码报错
+**A1**： 启动服务和发送请求时不要设置代理，可以在启动服务前和发送请求前关闭代理，关闭代理的命令是：
```
-python tools/serving/image_http_client.py 9292 ./docs/images/logo.png
+unset https_proxy
+unset http_proxy
```
-`9292` 为发送请求的端口号，需要与服务启动时的端口号保持一致，`./docs/images/logo.png` 为待识别的图像文件。最终返回 Top1 识别结果的类别 ID 以及概率值。
-
-* 更多的服务部署类型，如 `RPC预测服务` 等，可以参考 Serving 的 github 官网：[https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/imagenet](https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/imagenet)
+更多的服务部署类型，如 `RPC预测服务` 等，可以参考 Serving 的[github 官网](https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/imagenet)