hub 下载的 默认轻量级的 文字推理模型。有时候会很慢，有时候很快，运行时间越久，慢的越多。

[Nagisakurumi]

1.系统centos7 2.anaconda python3.7 3.环境paddleocr paddlehub 4.6核 i9 CPU 5.参数设置了8个线程 6.刚开始运行会比较快，但是时间运行久了就会变慢，这时候重启也没用。

[KPatr1ck]

你好，具体使用的是paddlehub的版本号是多少？ 用到的module和具体执行的脚本能顺便提供一下吗？

[Nagisakurumi]

paddlepaddle-gpu 2.0.0rc1.post110 虽然用的GPU版本的 但是实际运行的是 CPU 模式

from paddleocr import PaddleOCR import os import paddlehub as hub from paddlehub.common.logger import logger from paddlehub.module.module import moduleinfo, runnable, serving import datetime import sys import base64 import cv2 as cv import numpy as np from PIL import Image

def image_to_base64(image_path):

image_np = cv.imread(image_path)

image_np = np.array(Image.open(image_path))
# print(image_np)
image = cv.imencode('.jpg', image_np)[1]
image_code = str(base64.b64encode(image))[2:-1]
return image_code

def appendPath(path): current = os.getcwd() current = os.path.join(current, path) return current

def parse_args(): import argparse

return argparse.Namespace(
        use_gpu=False,
        ir_optim=True,
        use_tensorrt=False,
        gpu_mem=8000,
        image_dir='',
        det_algorithm='DB',
        det_model_dir=None,
        det_limit_side_len=960,
        det_limit_type='max',
        det_db_thresh=0.3,
        det_db_box_thresh=0.5,
        det_db_unclip_ratio=2.0,
        det_east_score_thresh=0.8,
        det_east_cover_thresh=0.1,
        det_east_nms_thresh=0.2,
        rec_algorithm='CRNN',
        rec_model_dir=None,
        rec_image_shape="3, 32, 320",
        rec_char_type='ch',
        rec_batch_num=30,
        max_text_length=25,
        rec_char_dict_path=appendPath('ppocr_keys_v1.txt'),
        use_space_char=True,
        drop_score=0.5,
        cls_model_dir=None,
        cls_image_shape="3, 48, 192",
        label_list=['0', '180'],
        cls_batch_num=30,
        cls_thresh=0.9,
        enable_mkldnn=False,
        use_zero_copy_run=False,
        use_pdserving=False,
        lang='ch',
        det=True,
        rec=True,
        use_angle_cls=False)

@moduleinfo( name="searchAll", version="1.0.0", summary="ocr system service", author="paddle-dev", author_email="paddle-dev@baidu.com", type="cv/text_recognition") class OcrSystem(hub.Module):

def init(self):

#     self.ocr_engine = PaddleOCR(**(parse_args().__dict__))

def _initialize(self, use_gpu=False, enable_mkldnn=False):
    self.ocr_engine = PaddleOCR(**(parse_args().__dict__))

#base64 -> image
def base64_to_image(self, base64_code):
    # base64解码
    img_data = base64.b64decode(base64_code)
    # 转换为np数组
    img_array = np.fromstring(img_data, np.uint8)
    # 转换成opencv可用格式
    img_array = cv.imdecode(img_array, cv.COLOR_RGB2BGR)

    return img_array

def convertToRect(self, points):
    x1 = min(points[0][0], points[1][0], points[2][0], points[3][0])
    y1 = min(points[0][1], points[1][1], points[2][1], points[3][1])

    x2 = max(points[0][0], points[1][0], points[2][0], points[3][0])
    y2 = max(points[0][1], points[1][1], points[2][1], points[3][1])

    return {'x' : int(x1), 'y' : int(y1), 'width' : int(x2 - x1), 'height' : int(y2 - y1)}

#查找所有文字
def searchAllProcess(self, data):
    img = self.base64_to_image(data)
    results = self.ocr_engine.ocr(img)
    if len(results) == 0:
        ret = []
        return ret
    boxes = []
    boxes = [{'text' : item[1][0], 'rect' : self.convertToRect(item[0])} for item in results]
    return boxes

@serving
def searchAll(self, data, **kwargs):
    try:
        start = datetime.datetime.now()
        if data is None or data == '':
            ret = []
            return ret

        ret = self.searchAllProcess(data)
        end = datetime.datetime.now()
        print("服务响应时间为: {} ms".format((end - start).microseconds / 1000.0))
        return ret
    except Exception as e:
        logger.info(e)
        return []

脚本就是一直发 jpg的图片

[KPatr1ck]

是用hub serving起了推理的服务，然后另一边一直发送图片去预测吗？

[Nagisakurumi]

是的，但是频率没有很高，