ubuntu上安装PaddleOCR环境(三)
基于HubServing的部署
1. 准备环境
# 安装paddlehub
pip3 install paddlehub --upgrade -i https://pypi.tuna.tsinghua.edu.cn/simple
# 设置环境变量
export PYTHONPATH=.
2. 安装服务模块
PaddleOCR提供3种服务模块,根据需要安装所需模块。如:
安装检测服务模块:
hub install deploy/hubserving/ocr_det/
或,安装识别服务模块:
hub install deploy/hubserving/ocr_rec/
或,安装检测+识别串联服务模块:
hub install deploy/hubserving/ocr_system/
3.配置文件启动(支持CPU、GPU)
其中,config.json格式如下:
{
"modules_info": {
"ocr_system": {
"init_args": {
"version": "1.0.0",
"use_gpu": true
},
"predict_args": {
}
}
},
"port": 8868,
"use_multiprocess": false,
"workers": 2
}
4.添加inference模型
下载超轻量级中文检测模型:
wget -P ./ch_lite/ https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db.tar && tar xf ./ch_lite/ch_det_mv3_db.tar -C ./ch_lite/
上述模型是以MobileNetV3为backbone训练的DB算法,将训练好的模型转换成inference模型只需要运行如下命令:
# -c后面设置训练算法的yml配置文件
# -o配置可选参数
# Global.checkpoints参数设置待转换的训练模型地址,不用添加文件后缀.pdmodel,.pdopt或.pdparams。
# Global.save_inference_dir参数设置转换的模型将保存的地址。
python3 tools/export_model.py -c configs/det/det_mv3_db.yml -o Global.checkpoints=./ch_lite/det_mv3_db/best_accuracy Global.save_inference_dir=./inference/det_db/
识别模型转inference模型
下载超轻量中文识别模型:
wget -P ./ch_lite/ https://paddleocr.bj.bcebos.com/ch_models/ch_rec_mv3_crnn.tar && tar xf ./ch_lite/ch_rec_mv3_crnn.tar -C ./ch_lite/
识别模型转inference模型与检测的方式相同,如下:
# -c后面设置训练算法的yml配置文件
# -o配置可选参数
# Global.checkpoints参数设置待转换的训练模型地址,不用添加文件后缀.pdmodel,.pdopt或.pdparams。
# Global.save_inference_dir参数设置转换的模型将保存的地址。
python3 tools/export_model.py -c configs/rec/rec_chinese_lite_train.yml -o Global.checkpoints=./ch_lite/rec_mv3_crnn/best_accuracy \
Global.save_inference_dir=./inference/rec_crnn/
5.修改配置
在deploy/hubserving/ocr_system/params.py中修改模型路径参数det_model_dir和rec_model_dir以及rec_char_dict_path的路径
6.在deploy/hubserving/ocr_system/module.py的from tools.infer.utility import base64_to_cv2前添加
import sys
sys.path.append('../../../')
避免找不到tooles模块的错误
7.启动hub
export CUDA_VISIBLE_DEVICES=0
hub serving start -c deploy/hubserving/ocr_system/config.json
8.测试:
官方测试:python tools/test_hubserving.py http://127.0.0.1:8868/predict/ocr_system ./doc/imgs/
我是在座面上修改的test_hubserving.py
#!/usr/bin/env python
#-*- coding:UTF-8 -*-(添加
# 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 os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(__dir__)
sys.path.append(os.path.abspath(os.path.join(__dir__, '..')))
import cv2
import numpy as np
import time
from PIL import Image, ImageDraw, ImageFont
import requests
import json
import base64
import math
def cv2_to_base64(image):
return base64.b64encode(image).decode('utf8')
def str_count(s):
"""
Count the number of Chinese characters,
a single English character and a single number
equal to half the length of Chinese characters.
args:
s(string): the input of string
return(int):
the number of Chinese characters
"""
import string
count_zh = count_pu = 0
s_len = len(s)
en_dg_count = 0
for c in s:
if c in string.ascii_letters or c.isdigit() or c.isspace():
en_dg_count += 1
elif c.isalpha():
count_zh += 1
else:
count_pu += 1
return s_len - math.ceil(en_dg_count / 2)
def text_visual(texts, scores, img_h=400, img_w=600, threshold=0.):
"""
create new blank img and draw txt on it
args:
texts(list): the text will be draw
scores(list|None): corresponding score of each txt
img_h(int): the height of blank img
img_w(int): the width of blank img
return(array):
"""
if scores is not None:
assert len(texts) == len(
scores), "The number of txts and corresponding scores must match"
def create_blank_img():
blank_img = np.ones(shape=[img_h, img_w], dtype=np.int8) * 255
blank_img[:, img_w - 1:] = 0
blank_img = Image.fromarray(blank_img).convert("RGB")
draw_txt = ImageDraw.Draw(blank_img)
return blank_img, draw_txt
blank_img, draw_txt = create_blank_img()
font_size = 20
txt_color = (0, 0, 0)
font = ImageFont.truetype("./doc/simfang.ttf", font_size, encoding="utf-8")
gap = font_size + 5
txt_img_list = []
count, index = 1, 0
for idx, txt in enumerate(texts):
index += 1
if scores[idx] < threshold or math.isnan(scores[idx]):
index -= 1
continue
first_line = True
while str_count(txt) >= img_w // font_size - 4:
tmp = txt
txt = tmp[:img_w // font_size - 4]
if first_line:
new_txt = str(index) + ': ' + txt
first_line = False
else:
new_txt = ' ' + txt
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
txt = tmp[img_w // font_size - 4:]
if count >= img_h // gap - 1:
txt_img_list.append(np.array(blank_img))
blank_img, draw_txt = create_blank_img()
count = 0
count += 1
if first_line:
new_txt = str(index) + ': ' + txt + ' ' + '%.3f' % (scores[idx])
else:
new_txt = " " + txt + " " + '%.3f' % (scores[idx])
draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
# whether add new blank img or not
if count >= img_h // gap - 1 and idx + 1 < len(texts):
txt_img_list.append(np.array(blank_img))
blank_img, draw_txt = create_blank_img()
count = 0
count += 1
txt_img_list.append(np.array(blank_img))
if len(txt_img_list) == 1:
blank_img = np.array(txt_img_list[0])
else:
blank_img = np.concatenate(txt_img_list, axis=1)
return np.array(blank_img)
def resize_img(img, input_size=600):
"""
resize img and limit the longest side of the image to input_size
"""
img = np.array(img)
im_shape = img.shape
im_size_max = np.max(im_shape[0:2])
im_scale = float(input_size) / float(im_size_max)
im = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
return im
def draw_ocr(image, boxes, txts, scores, draw_txt=True, drop_score=0.5):
"""
Visualize the results of OCR detection and recognition
args:
image(Image|array): RGB image
boxes(list): boxes with shape(N, 4, 2)
txts(list): the texts
scores(list): txxs corresponding scores
draw_txt(bool): whether draw text or not
drop_score(float): only scores greater than drop_threshold will be visualized
return(array):
the visualized img
"""
if scores is None:
scores = [1] * len(boxes)
for (box, score) in zip(boxes, scores):
if score < drop_score or math.isnan(score):
continue
box = np.reshape(np.array(box), [-1, 1, 2]).astype(np.int64)
image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
if draw_txt:
img = np.array(resize_img(image, input_size=1000))
txt_img = text_visual(
txts, scores, img_h=img.shape[0], img_w=600, threshold=drop_score)
img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
return img
return image
def draw_boxes(image, boxes, scores=None, drop_score=0.5):
if scores is None:
scores = [1] * len(boxes)
for (box, score) in zip(boxes, scores):
if score < drop_score:
continue
box = np.reshape(np.array(box), [-1, 1, 2]).astype(np.int64)
image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
return image
def draw_server_result(image_file, res):
img = cv2.imread(image_file)
image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
if len(res) == 0:
return np.array(image)
keys = res[0].keys()
if 'text_region' not in keys: # for ocr_rec, draw function is invalid
print("draw function is invalid for ocr_rec!")
return None
elif 'text' not in keys: # for ocr_det
print("draw text boxes only!")
boxes = []
for dno in range(len(res)):
boxes.append(res[dno]['text_region'])
boxes = np.array(boxes)
draw_img = draw_boxes(image, boxes)
return draw_img
else: # for ocr_system
print("draw boxes and texts!")
boxes = []
texts = []
scores = []
for dno in range(len(res)):
boxes.append(res[dno]['text_region'])
texts.append(res[dno]['text'])
scores.append(res[dno]['confidence'])
boxes = np.array(boxes)
scores = np.array(scores)
draw_img = draw_ocr(image, boxes, texts, scores, draw_txt=True, drop_score=0.5)
return draw_img
def main(url, image_path):
is_visualize = True
headers = {"Content-type": "application/json"}
cnt = 0
total_time = 0
path =(r"C:\Users\EDZ\Desktop\img\\")
image_file = path+image_path
img = open(image_file, 'rb').read()
if img is None:
return
# 发送HTTP请求
starttime = time.time()
data = {'images':[cv2_to_base64(img)]}
r = requests.post(url=url, headers=headers, data=json.dumps(data))
elapse = time.time() - starttime
total_time += elapse
print("Predict time of %s: %.3fs" % (image_file, elapse))
res = r.json()["results"][0]
print(res)
if is_visualize:
draw_img = draw_server_result(image_file, res)
if draw_img is not None:
draw_img_save = path+"server_results\\"
if not os.path.exists(draw_img_save):
os.makedirs(draw_img_save)
cv2.imwrite(
os.path.join(draw_img_save, os.path.basename(image_file)),
draw_img[:, :, ::-1])
if __name__ == '__main__':
if len(sys.argv) != 3:
print("Usage: %s server_url image_path" % sys.argv[0])
else:
server_url = sys.argv[1]
image_path = sys.argv[2]
main(server_url, image_path)
ps:更改使用的模型需要重新装一次服务包
卸载旧服务包
hub uninstall ocr_system
安装修改后的新服务包
hub install deploy/hubserving/ocr_system/