前言
最近公司需要对图片中的不同的货车品牌和车系进行识别，通过PaddleClas进行模型训练后得到一个品牌识别模型和一个车系识别模型，现在对两个模型部署到一台华为云的GPU服务器上，要对多个模型同时进行部署，只能采取PaddleServing中的 Pipeline 服务或者C++ serving服务进行部署，由于C++ serving需要编译源码，比较麻烦，所以下面采用Pipeline 方式对多个模型进行串联部署。
如何在华为云服务器上搭建GPU版本的PaddlePaddle环境请参考以下文章： https://blog.csdn.net/loutengyuan/article/details/126527326
环境准备
需要准备PaddleClas的运行环境和Paddle Serving的运行环境。
- 准备PaddleClas的运行环境链接
# 克隆代码
git clone https://github.com/PaddlePaddle/PaddleClas
- 安装PaddleServing的运行环境，步骤如下
# 安装serving，用于启动服务
wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_server_gpu-0.8.3.post102-py3-none-any.whl
pip3 install paddle_serving_server_gpu-0.8.3.post102-py3-none-any.whl
# 安装client，用于向服务发送请求
wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_client-0.8.3-cp38-none-any.whl
pip3 install paddle_serving_client-0.8.3-cp38-none-any.whl
# 安装serving-app
wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_app-0.8.3-py3-none-any.whl
pip3 install paddle_serving_app-0.8.3-py3-none-any.whl
服务数据准备
使用 PaddleServing 做图像识别服务化部署时，需要将保存的多个 inference 模型都转换为 Serving 模型。
模型转换
进入工作目录：
cd PaddleClas/deploy/
创建并进入models文件夹：
# 创建并进入models文件夹
mkdir models
cd models
将训练好的inference 模型放到该文件夹下，包括检测模型（picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer）、品牌识别模型（rec_brands_v1.0_infer）和车系识别模型（rec_series_v1.0_infer）结构如下：
├── picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer
│ ├── infer_cfg.yml
│ ├── inference.pdiparams
│ ├── inference.pdiparams.info
│ └── inference.pdmodel
├── rec_brands_v1.0_infer
│ ├── inference.pdiparams
│ ├── inference.pdiparams.info
│ └── inference.pdmodel
└── rec_series_v1.0_infer
├── inference.pdiparams
├── inference.pdiparams.info
├── inference.pdmodel
└── readme.txt
转换通用检测 inference 模型为 Serving 模型：
# 转换通用检测模型
python3.8 -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/
通用检测 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，具备如下结构：
├── picodet_PPLCNet_x2_5_mainbody_lite_v1.0_serving/
│ ├── inference.pdiparams
│ ├── inference.pdmodel
│ ├── serving_server_conf.prototxt
│ └── serving_server_conf.stream.prototxt
│
└── picodet_PPLCNet_x2_5_mainbody_lite_v1.0_client/
├── serving_client_conf.prototxt
└── serving_client_conf.stream.prototxt
转换品牌识别 inference 模型为 Serving 模型：
# 转换品牌识别模型
python3.8 -m paddle_serving_client.convert \
--dirname ./rec_brands_v1.0_infer/ \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--serving_server ./rec_brands_v1.0_serving/ \
--serving_client ./rec_brands_v1.0_client/
品牌识别 inference 模型转换完成后，会在当前文件夹多出 rec_brands_v1.0_serving/ 和 rec_brands_v1.0_client/ 的文件夹，具备如下结构：
├── rec_brands_v1.0_serving/
│ ├── inference.pdiparams
│ ├── inference.pdmodel
│ ├── serving_server_conf.prototxt
│ └── serving_server_conf.stream.prototxt
│
└── rec_brands_v1.0_client/
├── serving_client_conf.prototxt
└── serving_client_conf.stream.prototxt
分别修改 rec_brands_v1.0_serving/ 和 rec_brands_v1.0_client/ 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: false
fetch_type: 1
shape: 512
}
转换车系识别 inference 模型为 Serving 模型：
# 转换车系识别模型
python3.8 -m paddle_serving_client.convert \
--dirname ./rec_series_v1.0_infer/ \
--model_filename inference.pdmodel \
--params_filename inference.pdiparams \
--serving_server ./rec_series_v1.0_serving/ \
--serving_client ./rec_series_v1.0_client/
车系识别 inference 模型转换完成后，会在当前文件夹多出 rec_series_v1.0_serving/ 和 rec_series_v1.0_client/ 的文件夹，具备如下结构：
├── rec_series_v1.0_serving/
│ ├── inference.pdiparams
│ ├── inference.pdmodel
│ ├── serving_server_conf.prototxt
│ └── serving_server_conf.stream.prototxt
│
└── rec_series_v1.0_client/
├── serving_client_conf.prototxt
└── serving_client_conf.stream.prototxt
分别修改 rec_series_v1.0_serving/ 和 rec_series_v1.0_client/ 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: false
fetch_type: 1
shape: 512
}
上述命令中参数具体含义如下表所示：
| 参数 | 类型 | 默认值 | 描述 |
|---|---|---|---|
dirname | str | - | 需要转换的模型文件存储路径，Program结构文件和参数文件均保存在此目录。 |
model_filename | str | None | 存储需要转换的模型Inference Program结构的文件名称。如果设置为None，则使用 __model__ 作为默认的文件名 |
params_filename | str | None | 存储需要转换的模型所有参数的文件名称。当且仅当所有模型参数被保>存在一个单独的二进制文件中，它才需要被指定。如果模型参数是存储在各自分离的文件中，设置它的值为None |
serving_server | str | "serving_server" | 转换后的模型文件和配置文件的存储路径。默认值为serving_server |
serving_client | str | "serving_client" | 转换后的客户端配置文件存储路径。默认值为serving_client |
索引库添加
将品牌和车系库放到上一级（deploy）目录
# 回到deploy目录
cd ../
目录结构如下：
├── brand_dataset_v1.0/
│ └── index
│ ├── id_map.pkl
│ └── vector.index
└── series_dataset_v1.0/
└── index
├── id_map.pkl
└── vector.index
服务部署
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
cd ./deploy/paddleserving/recognition
paddleserving 目录包含启动 Python Pipeline 服务、C++ Serving 服务和发送预测请求的代码，包括：
__init__.py
config.yml # 启动python pipeline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
readme.md # 识别模型服务化部署文档
run_cpp_serving.sh # 启动C++ Pipeline Serving部署的脚本
test_cpp_serving_client.py # rpc方式发送C++ Pipeline serving预测请求的脚本
修改config.yml文件如下：
#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: 1
#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: 8899
#rpc_port: 9994
dag:
#op资源类型, True, 为线程模型；False，为进程模型
is_thread_op: False
op:
rec_brands:
#并发数，is_thread_op=True时，为线程并发；否则为进程并发
concurrency: 1
#当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
local_service_conf:
#uci模型路径
model_config: ../../models/rec_brands_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
#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: "0" # "0,1"
#client类型，包括brpc, grpc和local_predictor.local_predictor不启动Serving服务，进程内预测
client_type: local_predictor
#Fetch结果列表，以client_config中fetch_var的alias_name为准
fetch_list: ["features"]
rec_series:
#并发数，is_thread_op=True时，为线程并发；否则为进程并发
concurrency: 1
#当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
local_service_conf:
#uci模型路径
model_config: ../../models/rec_series_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
#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: "0" # "0,1"
#client类型，包括brpc, grpc和local_predictor.local_predictor不启动Serving服务，进程内预测
client_type: local_predictor
#Fetch结果列表，以client_config中fetch_var的alias_name为准
fetch_list: ["features"]
det:
concurrency: 1
local_service_conf:
client_type: local_predictor
device_type: 1
devices: '0'
fetch_list:
- save_infer_model/scale_0.tmp_1
model_config: ../../models/picodet_PPLCNet_x2_5_mainbody_lite_v1.0_serving/
修改recognition_web_service.py文件如下：
# 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 datetime
from paddle_serving_server.web_service import WebService, Op
from paddle_serving_server.pipeline import RequestOp, ResponseOp
from paddle_serving_server.pipeline import PipelineServer
from paddle_serving_server.pipeline.proto import pipeline_service_pb2
from paddle_serving_server.pipeline.channel import ChannelDataErrcode
import logging
import numpy as np
import sys
import cv2
from paddle_serving_app.reader import *
import base64
import os
import faiss
import pickle
import json
class TestRequestOp(RequestOp):
def init_op(self):
pass
def unpack_request_package(self, request):
# print(str(request.method))
dict_data = {}
log_id = None
if request is None:
raise ValueError("request is None")
for idx, key in enumerate(request.key):
dict_data[key] = request.value[idx]
log_id = request.logid
return dict_data, log_id, None, ""
class DetOp(Op):
def init_op(self):
self.img_preprocess = Sequential([
BGR2RGB(), Div(255.0),
Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225], False),
Resize((640, 640)), Transpose((2, 0, 1))
])
self.img_postprocess = RCNNPostprocess("label_list.txt", "output")
self.threshold = 0.2
self.max_det_results = 5
def generate_scale(self, im):
"""
Args:
im (np.ndarray): image (np.ndarray)
Returns:
im_scale_x: the resize ratio of X
im_scale_y: the resize ratio of Y
"""
target_size = [640, 640]
origin_shape = im.shape[:2]
resize_h, resize_w = target_size
im_scale_y = resize_h / float(origin_shape[0])
im_scale_x = resize_w / float(origin_shape[1])
return im_scale_y, im_scale_x
def preprocess(self, input_dicts, data_id, log_id):
print("{} detect begin --> data_id: {}".format(datetime.datetime.now(), data_id))
(_, input_dict), = input_dicts.items()
imgs = []
raw_imgs = []
for key in input_dict.keys():
data = base64.b64decode(input_dict[key].encode('utf8'))
raw_imgs.append(data)
data = np.fromstring(data, np.uint8)
raw_im = cv2.imdecode(data, cv2.IMREAD_COLOR)
im_scale_y, im_scale_x = self.generate_scale(raw_im)
im = self.img_preprocess(raw_im)
im_shape = np.array(im.shape[1:]).reshape(-1)
scale_factor = np.array([im_scale_y, im_scale_x]).reshape(-1)
imgs.append({
"image": im[np.newaxis, :],
"im_shape": im_shape[np.newaxis, :],
"scale_factor": scale_factor[np.newaxis, :],
})
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)
}
return feed_dict, False, None, ""
def postprocess(self, input_dicts, fetch_dict, data_id, 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 = 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}
print("{} detect finish --> data_id: {}".format(datetime.datetime.now(), data_id))
return res_dict, None, ""
class BrandsRecOp(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))
])
index_dir = "../../brand_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)
self.rec_nms_thresold = 0.05
self.rec_score_thres = 0.5
self.feature_normalize = True
self.return_k = 1
self.area_ratio_thresold=0.1
def preprocess(self, input_dicts, data_id, log_id):
(_, input_dict), = input_dicts.items()
raw_img = input_dict["image"][0]
data = np.frombuffer(raw_img, np.uint8)
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]]
})
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()
img = self.seq(im)
imgs.append(img[np.newaxis, :].copy())
input_imgs = np.concatenate(imgs, axis=0)
return {"x": input_imgs}, False, None, ""
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")
x2 = np.array([r["bbox"][2] for r in results]).astype("float32")
y2 = np.array([r["bbox"][3] for r in results]).astype("float32")
scores = np.array([r["rec_scores"] for r in results])
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1]
while order.size > 0:
i = order[0]
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= thresh)[0]
order = order[inds + 1]
filtered_results.append(results[i])
return filtered_results
def check_boxes(self, results, area_ratio_thresh=0.1):
filtered_results = []
for result in results:
if result["area_ratio"]>=area_ratio_thresh:
filtered_results.append(result)
if len(filtered_results)>0:
return filtered_results
else:
return results
def postprocess(self, input_dicts, fetch_dict, data_id, log_id):
batch_features = fetch_dict["features"]
if self.feature_normalize:
feas_norm = np.sqrt(
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)
origin_img_box = self.det_boxes[len(self.det_boxes) - 1]["bbox"]
total_pixes = origin_img_box[2] * origin_img_box[3]
results = []
for i in range(scores.shape[0]):
pred = {}
xmin, ymin, xmax, ymax = self.det_boxes[i]["bbox"]
area_pix = (xmax - xmin) * (ymax - ymin)
ratio = 0.0
if total_pixes > 0:
ratio = area_pix * 1.0 / total_pixes
if scores[i][0] >= self.rec_score_thres:
pred["bbox"] = [int(x) for x in self.det_boxes[i]["bbox"]]
pred["rec_docs"] = self.id_map[docs[i][0]].split()[1]
pred["rec_scores"] = scores[i][0]
pred["area_ratio"] = round(ratio, 4)
results.append(pred)
#do nms
results = self.nms_to_rec_results(results, self.rec_nms_thresold)
print("{} BrandsRecOp data_id: {} --> Nms Result: {}".format(datetime.datetime.now(), data_id, results))
results = self.check_boxes(results, self.area_ratio_thresold)
print("{} BrandsRecOp data_id: {} --> Out Result: {}".format(datetime.datetime.now(), data_id, results))
return {"result": str(results)}, None, ""
class SeriesRecOp(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))
])
index_dir = "../../series_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)
self.rec_nms_thresold = 0.05
self.rec_score_thres = 0.5
self.feature_normalize = True
self.return_k = 1
self.area_ratio_thresold=0.1
def preprocess(self, input_dicts, data_id, log_id):
(_, input_dict), = input_dicts.items()
raw_img = input_dict["image"][0]
data = np.frombuffer(raw_img, np.uint8)
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]]
})
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()
img = self.seq(im)
imgs.append(img[np.newaxis, :].copy())
input_imgs = np.concatenate(imgs, axis=0)
return {"x": input_imgs}, False, None, ""
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")
x2 = np.array([r["bbox"][2] for r in results]).astype("float32")
y2 = np.array([r["bbox"][3] for r in results]).astype("float32")
scores = np.array([r["rec_scores"] for r in results])
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1]
while order.size > 0:
i = order[0]
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= thresh)[0]
order = order[inds + 1]
filtered_results.append(results[i])
return filtered_results
def check_boxes(self, results, area_ratio_thresh=0.1):
filtered_results = []
for result in results:
if result["area_ratio"]>=area_ratio_thresh:
filtered_results.append(result)
if len(filtered_results)>0:
return filtered_results
else:
return results
def postprocess(self, input_dicts, fetch_dict, data_id, log_id):
batch_features = fetch_dict["features"]
if self.feature_normalize:
feas_norm = np.sqrt(
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)
origin_img_box = self.det_boxes[len(self.det_boxes) - 1]["bbox"]
total_pixes = origin_img_box[2] * origin_img_box[3]
results = []
for i in range(scores.shape[0]):
pred = {}
xmin, ymin, xmax, ymax = self.det_boxes[i]["bbox"]
area_pix = (xmax - xmin) * (ymax - ymin)
ratio = 0.0
if total_pixes > 0:
ratio = area_pix * 1.0 / total_pixes
if scores[i][0] >= self.rec_score_thres:
pred["bbox"] = [int(x) for x in self.det_boxes[i]["bbox"]]
pred["rec_docs"] = self.id_map[docs[i][0]].split()[1]
pred["rec_scores"] = scores[i][0]
pred["area_ratio"] = round(ratio, 4)
results.append(pred)
#do nms
results = self.nms_to_rec_results(results, self.rec_nms_thresold)
print("{} SeriesRecOp data_id: {} --> Nms Result: {}".format(datetime.datetime.now(), data_id, results))
results = self.check_boxes(results, self.area_ratio_thresold)
print("{} SeriesRecOp data_id: {} --> Out Result: {}".format(datetime.datetime.now(), data_id, results))
return {"result": str(results)}, None, ""
class CombineOp(Op):
def preprocess(self, input_data, data_id, log_id):
return None, False, None, ""
def postprocess(self, input_dicts, fetch_dict, data_id, log_id):
print("{} CombineOp data_id: {} --> input_dicts: {}".format(datetime.datetime.now(), data_id, input_dicts))
results = {}
for op_name, data in input_dicts.items():
if "brands" in op_name:
ret = data["result"]
if ret is not None:
results["brands"] = json.loads(ret.replace("'", "\""))
else:
results["brands"] = "[]"
elif "series" in op_name:
ret = data["result"]
if ret is not None:
results["series"] = json.loads(ret.replace("'", "\""))
else:
results["series"] = "[]"
print("{} CombineOp data_id: {} --> Out Result: {}".format(datetime.datetime.now(), data_id, results))
return {"result": str(results)}, None, ""
class RecognitionService(WebService):
def get_pipeline_response(self, read_op):
read_op2 = TestRequestOp()
det_op = DetOp(name="det", input_ops=[read_op2])
rec_brands_op = BrandsRecOp(name="rec_brands", input_ops=[det_op])
rec_series_op = SeriesRecOp(name="rec_series", input_ops=[det_op])
combine_op = CombineOp("combine", input_ops=[rec_brands_op, rec_series_op])
return combine_op
product_recog_service = RecognitionService(name="recognition")
product_recog_service.prepare_pipeline_config("config.yml")
product_recog_service.run_service()
启动服务：
# 启动服务，运行日志保存在 log.txt
nohup python3.8 recognition_web_service.py &>log.txt &
如果出现faiss没找到，请参考这里：https://blog.csdn.net/weixin_43882112/article/details/107614217
查看进程
ps -ef|grep python
关闭进程
kill -9 19913
查看日志
tail -f 1000 log.log
如何查看端口占用
$: netstat -anp | grep 8888
tcp 0 0 127.0.0.1:8888 0.0.0.0:* LISTEN 13404/python3
tcp 0 1 172.17.0.10:34036 115.42.35.84:8888 SYN_SENT 14586/python3
强制杀掉进程：通过pid
$: kill -9 13404
$: kill -9 14586
$: netstat -anp | grep 8888
$:
服务测试
修改pipeline_http_client.py文件如下：
import requests
import json
import base64
import os
imgpath = "图片路径.jpg"
def cv2_to_base64(image):
return base64.b64encode(image).decode('utf8')
if __name__ == "__main__":
url = "http://127.0.0.1:8899/recognition/prediction"
with open(os.path.join(".", imgpath), 'rb') as file:
image_data1 = file.read()
image = cv2_to_base64(image_data1)
data = {"key": ["image"], "value": [image]}
for i in range(1):
r = requests.post(url=url, data=json.dumps(data))
print(r.json())
发送请求：
python3.8 pipeline_http_client.py
成功运行后，模型预测的结果会打印在客户端中，如下所示：
{'err_no': 0, 'err_msg': '', 'key': ['result'], 'value': ["{'brands': [{'bbox': [16, 19, 492, 565], 'rec_docs': '1', 'rec_scores': 0.98805684, 'area_ratio': 0.7432}], 'series': [{'bbox': [16, 19, 492, 565], 'rec_docs': '6', 'rec_scores': 0.9267364, 'area_ratio': 0.7432}]}"], 'tensors': []}
扫一扫
3万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
