项目架构调整，模型全部采用接口调用

services/paddle_services/Dockerfile

@@ -0,0 +1,29 @@
+# 使用官方的paddle镜像作为基础
+FROM registry.baidubce.com/paddlepaddle/paddle:2.6.1-gpu-cuda12.0-cudnn8.9-trt8.6
+
+# 设置工作目录
+WORKDIR /app
+
+# 设置环境变量
+ENV PYTHONUNBUFFERED=1 \
+    # 设置时区
+    TZ=Asia/Shanghai \
+    # 设置pip镜像地址，加快安装速度
+    PIP_INDEX_URL=https://pypi.tuna.tsinghua.edu.cn/simple
+
+# 安装依赖
+COPY requirements.txt /app/requirements.txt
+COPY packages /app/packages
+RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo '$TZ' > /etc/timezone \
+    && pip install --no-cache-dir -r requirements.txt \
+    && pip uninstall -y onnxruntime onnxruntime-gpu \
+    && pip install onnxruntime-gpu==1.18.0 --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/
+
+# 将当前目录内容复制到容器的/app内
+COPY . /app
+
+# 暴露端口
+# EXPOSE 8081
+
+# 运行api接口，具体接口在命令行或docker-compose.yml文件中定义
+ENTRYPOINT ["gunicorn"]

services/paddle_services/doc_dewarp/__init__.py

@@ -1,4 +1,7 @@
+import os.path
+
 from onnxruntime import InferenceSession
 
-DOC_TR = InferenceSession("model/dewarp_model/doc_tr_pp.onnx",
-                          providers=["CUDAExecutionProvider"], provider_options=[{"device_id": 0}])
+MODEL_PATH = os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))),
+                          'model', 'dewarp_model', 'doc_tr_pp.onnx')
+DOC_TR = InferenceSession(MODEL_PATH, providers=['CUDAExecutionProvider'], provider_options=[{'device_id': 0}])

services/paddle_services/doc_dewarp/dewarper.py

@@ -11,10 +11,10 @@ def dewarp_image(image):
     y = to_tensor(image)
 
     img = np.transpose(img, (2, 0, 1))
-    bm = DOC_TR.run(None, {"image": img[None,]})[0]
+    bm = DOC_TR.run(None, {'image': img[None,]})[0]
     bm = paddle.to_tensor(bm)
     bm = paddle.nn.functional.interpolate(
-        bm, y.shape[2:], mode="bilinear", align_corners=False
+        bm, y.shape[2:], mode='bilinear', align_corners=False
     )
     bm_nhwc = np.transpose(bm, (0, 2, 3, 1))
     out = paddle.nn.functional.grid_sample(y, (bm_nhwc / 288 - 0.5) * 2)

services/paddle_services/paddle_detection/__init__.py

@@ -1,4 +1,8 @@
+import os
+
 from onnxruntime import InferenceSession
 
-PADDLE_DET = InferenceSession("model/object_det_model/ppyoloe_plus_crn_l_80e_coco_w_nms.onnx",
-                              providers=["CPUExecutionProvider"], provider_options=[{"device_id": 0}])
+MODEL_DIR = os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))),
+                         'model', 'object_det_model')
+PADDLE_DET = InferenceSession(os.path.join(MODEL_DIR, 'ppyoloe_plus_crn_l_80e_coco_w_nms.onnx'),
+                              providers=['CPUExecutionProvider'], provider_options=[{'device_id': 0}])

services/paddle_services/paddle_detection/detector.py

@@ -1,13 +1,13 @@
-import tempfile
+import os.path
 from collections import defaultdict
 
 import cv2
 import numpy as np
 
-from paddle_detection import PADDLE_DET
-from paddle_detection.deploy.third_engine.onnx.infer import PredictConfig
-from paddle_detection.deploy.third_engine.onnx.preprocess import Compose
-from util import image_util, common_util
+from util import image_util
+from . import PADDLE_DET, MODEL_DIR
+from .deploy.third_engine.onnx.infer import PredictConfig
+from .deploy.third_engine.onnx.preprocess import Compose
 
 
 def predict_image(infer_config, predictor, img_path):
@@ -15,7 +15,7 @@ def predict_image(infer_config, predictor, img_path):
     transforms = Compose(infer_config.preprocess_infos)
     # predict image
     inputs = transforms(img_path)
-    inputs["image"] = np.array(inputs["image"]).astype('float32')
+    inputs['image'] = np.array(inputs['image']).astype('float32')
     inputs_name = [var.name for var in predictor.get_inputs()]
     inputs = {k: inputs[k][None,] for k in inputs_name}
 
@@ -25,25 +25,23 @@ def predict_image(infer_config, predictor, img_path):
     result = defaultdict(list)
     for bbox in bboxes:
         if bbox[0] > -1 and bbox[1] > infer_config.draw_threshold:
-            result[bbox[0]].append({"score": bbox[1], "box": bbox[2:]})
+            result[bbox[0]].append({'score': bbox[1], 'box': bbox[2:]})
     return result
 
 
 def detect_image(img_path):
-    infer_cfg = "model/object_det_model/infer_cfg.yml"
+    infer_cfg = os.path.join(MODEL_DIR, 'infer_cfg.yml')
     # load infer config
     infer_config = PredictConfig(infer_cfg)
 
     return predict_image(infer_config, PADDLE_DET, img_path)
 
 
-def get_book_areas(image):
-    with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
-        cv2.imwrite(temp_file.name, image)
-    detect_result = detect_image(temp_file.name)
-    common_util.delete_temp_file(temp_file.name)
+def get_book_areas(img_path):
+    detect_result = detect_image(img_path)
     book_areas = detect_result[73]
     result = []
+    image = cv2.imread(img_path)
     for book_area in book_areas:
-        result.append(image_util.capture(image, book_area["box"]))
+        result.append(image_util.capture(image, book_area['box']))
     return result

services/paddle_services/requestments.txt

@@ -0,0 +1,16 @@
+numpy==1.26.4
+onnxconverter-common==1.14.0
+OpenCC==1.1.6
+opencv-python==4.6.0.66
+paddle2onnx==1.2.3
+paddleclas==2.5.2
+paddlenlp==2.6.1
+paddleocr==2.7.3
+pillow==10.4.0
+pymysql==1.1.1
+requests==2.32.3
+sqlacodegen==2.3.0.post1
+sqlalchemy==1.4.52
+tenacity==8.5.0
+ufile==3.2.9
+zxing-cpp==2.2.0