📄 Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.
update rapidocr_api (#242)
xmxoxo3aa4463ad20b
diff --git a/api/Dockerfile b/api/Dockerfilenew file mode 100644index 0000000..9a0ba26--- /dev/null+++ b/api/Dockerfile@@ -0,0 +1,18 @@+FROM python:3.10.11-slim-buster++ENV DEBIAN_FRONTEND=noninteractive++# 设置工作目录+WORKDIR /app++RUN pip install --no-cache-dir rapidocr_api -i https://mirrors.aliyun.com/pypi/simple++RUN pip uninstall -y opencv-python && \+ pip install --no-cache-dir opencv-python-headless -i https://mirrors.aliyun.com/pypi/simple++EXPOSE 9003++# 升级后可用+# CMD ["bash", "-c", "rapidocr_api -ip 0.0.0.0 -p 9003 -workers 2"]+CMD ["rapidocr_api"]+diff --git a/api/README.md b/api/README.mdindex 3d153be..367fa02 100644--- a/api/README.md+++ b/api/README.md@@ -1 +1,54 @@### See [Documentation](https://rapidai.github.io/RapidOCRDocs/install_usage/rapidocr_api/usage/)+++### API 修改说明++* uvicorn启动时,reload参数设置为False,避免反复加载;+* 增加了启动参数: workers,可启动多个实例,以满足多并发需求。+* 可通过环境变量传递模型参数:det_model_path, cls_model_path, rec_model_path;+* 接口中可传入参数,控制是否使用检测、方向分类和识别这三部分的模型;客户端调用见`demo.py`+* 增加了Dockerfile,可自行构建镜像。++启动服务端:++Windows下启动:++```for win shell+set det_model_path=I:\models\图像相关\OCR\RapidOCR\PP-OCRv4\ch_PP-OCRv4_det_server_infer.onnx+set det_model_path=++set rec_model_path=I:\models\图像相关\OCR\RapidOCR\PP-OCRv4\ch_PP-OCRv4_rec_server_infer.onnx+rapidocr_api+```++Linux下启动:+```shell+# 默认参数启动+rapidocr_api++# 指定参数:端口与进程数量;+rapidocr_api -ip 0.0.0.0 -p 9005 -workers 2++# 指定模型+expert det_model_path=/mnt/sda1/models/PP-OCRv4/ch_PP-OCRv4_det_server_infer.onnx+expert rec_model_path=/mnt/sda1/models/PP-OCRv4/ch_PP-OCRv4_rec_server_infer.onnx+rapidocr_api -ip 0.0.0.0 -p 9005 -workers 2+```+++客户端调用说明:+```+cd api+python demo.py+```++构建镜像:+```+cd api+sudo docker build -t="rapidocr_api:0.1.1" .+```+启动镜像:++```+docker run -p 9003:9003 --name rapidocr_api1 --restart always -d rapidocr_api:0.1.1+```\ No newline at end of filediff --git a/api/demo.py b/api/demo.pyindex 865292d..cbf1600 100644--- a/api/demo.py+++ b/api/demo.py@@ -13,17 +13,49 @@# print(response.text)+import timeimport base64-import requestsurl = "http://localhost:9003/ocr"img_path = "../python/tests/test_files/ch_en_num.jpg"+# 方式一:使用base64编码传+stime = time.time()with open(img_path, "rb") as fa:img_str = base64.b64encode(fa.read())payload = {"image_data": img_str}-response = requests.post(url, data=payload, timeout=60)+response = requests.post(url, data=payload) #, timeout=60print(response.json())+etime = time.time() - stime+print(f'用时:{etime:.3f}秒')++print('-'*40)++# 方式二:使用文件上传方式+stime = time.time()+with open(img_path, 'rb') as f:+ file_dict = {'image_file': (img_path, f, 'image/png')}+ response = requests.post(url, files=file_dict) #, timeout=60+ print(response.json())++etime = time.time() - stime+print(f'用时:{etime:.3f}秒')+print('-'*40)++# 方式三:控制是否使用检测、方向分类和识别这三部分的模型; 不使用检测模型:use_det=False+stime = time.time()+img_path = "../python/tests/test_files/test_without_det.jpg"++with open(img_path, 'rb') as f:+ file_dict = {'image_file': (img_path, f, 'image/png')}+ # 添加控制参数+ data = {"use_det":False, "use_cls":True, "use_rec":True}+ response = requests.post(url, files=file_dict, data=data) #, timeout=60+ print(response.json())++etime = time.time() - stime+print(f'用时:{etime:.3f}秒')+print('-'*40)diff --git a/api/rapidocr_api/main.py b/api/rapidocr_api/main.pyindex 5ac113c..7c3a2c3 100644--- a/api/rapidocr_api/main.py+++ b/api/rapidocr_api/main.py@@ -1,9 +1,11 @@# -*- encoding: utf-8 -*-# @Author: SWHL# @Contact: [email protected]+import argparseimport base64import io+import osimport sysfrom pathlib import Pathfrom typing import Dict@@ -19,37 +21,47 @@ sys.path.append(str(Path(__file__).resolve().parent.parent))class OCRAPIUtils:def __init__(self) -> None:- self.ocr = RapidOCR()+ # 从环境变量中读取参数+ det_model_path = os.getenv("det_model_path", None)+ cls_model_path = os.getenv("cls_model_path", None)+ rec_model_path = os.getenv("rec_model_path", None)++ self.ocr = RapidOCR(det_model_path=det_model_path, cls_model_path=cls_model_path, rec_model_path=rec_model_path)- def __call__(self, img: Image.Image) -> Dict:+ def __call__(self, img: Image.Image, use_det=None, use_cls=None, use_rec=None) -> Dict:img = np.array(img)- ocr_res, _ = self.ocr(img)+ ocr_res, _ = self.ocr(img, use_det=use_det, use_cls=use_cls, use_rec=use_rec)if not ocr_res:return {}- out_dict = {- str(i): {- "rec_txt": rec,- "dt_boxes": dt_box,- "score": f"{score:.4f}",- }- for i, (dt_box, rec, score) in enumerate(ocr_res)- }+ # 转换为字典格式: 兼容所有参数情况+ out_dict = {}+ for i, dats in enumerate(ocr_res):+ values = {}+ for dat in dats:+ if type(dat) == str:+ values["rec_txt"] = dat+ if type(dat) == np.float64:+ values["score"] = f"{dat:.4f}"+ if type(dat) == list:+ values["dt_boxes"] = dat+ out_dict[str(i)] = values+return out_dictapp = FastAPI()processor = OCRAPIUtils()-@app.get("/")async def root():return {"message": "Welcome to RapidOCR API Server!"}-@app.post("/ocr")-async def ocr(image_file: UploadFile = None, image_data: str = Form(None)):+async def ocr(image_file: UploadFile = None, image_data: str = Form(None),+ use_det: bool = Form(None), use_cls: bool = Form(None), use_rec: bool = Form(None)):+if image_file:img = Image.open(image_file.file)elif image_data:@@ -60,19 +72,17 @@ async def ocr(image_file: UploadFile = None, image_data: str = Form(None)):raise ValueError("When sending a post request, data or files must have a value.")-- ocr_res = processor(img)+ ocr_res = processor(img, use_det=use_det, use_cls=use_cls, use_rec=use_rec)return ocr_res-def main():parser = argparse.ArgumentParser("rapidocr_api")parser.add_argument("-ip", "--ip", type=str, default="0.0.0.0", help="IP Address")parser.add_argument("-p", "--port", type=int, default=9003, help="IP port")+ parser.add_argument('-workers', "--workers", type=int, default=1, help='number of worker process')args = parser.parse_args()- uvicorn.run("rapidocr_api.main:app", host=args.ip, port=args.port, reload=True)-+ uvicorn.run("rapidocr_api.main:app", host=args.ip, port=args.port, reload=0, workers=args.workers)if __name__ == "__main__":main()