优化文档检测的结果

docker-compose.yml

@@ -1,8 +1,8 @@
 x-env:
   &template
-  # 启动时可以指定环境变量TAG来手动选择镜像版本
-  # docker-compose up -d -e TAG=1.0.0
-  image: fcb_photo_review:${TAG:-1.13.8}
+  # 可以在项目中创建.env文件，在其中定义TAG的值，使用以下命令在启动时替代默认的TAG
+  # docker-compose --env-file .env up -d
+  image: fcb_photo_review:${TAG:-1.13.9}
   restart: always
 
 services:
@@ -132,6 +132,24 @@ services:
               capabilities: [ "gpu" ]
               driver: "nvidia"
 
+  photo_review_7:
+    <<: *template
+    container_name: photo_review_7
+    hostname: photo_review_7
+    volumes:
+      - ./log:/app/log
+      - ./model:/app/model
+    depends_on:
+      - photo_review_6
+    command: [ "photo_review.py" ]
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - device_ids: [ "0", "1" ]
+              capabilities: [ "gpu" ]
+              driver: "nvidia"
+
   photo_mask_1:
     <<: *template
     container_name: photo_mask_1
@@ -139,7 +157,7 @@ services:
     volumes:
       - ./log:/app/log
     depends_on:
-      - photo_review_6
+      - photo_review_7
     command: [ "photo_mask.py", "--clean", "True" ]
     deploy:
       resources:

paddle_detection/__init__.py

@@ -1,4 +1,4 @@
 from onnxruntime import InferenceSession
 
-PADDLE_DET = InferenceSession("model/object_det_model/ppyoloe_plus_crn_x_80e_coco_w_nms.onnx",
+PADDLE_DET = InferenceSession("model/object_det_model/ppyoloe_plus_crn_l_80e_coco_w_nms.onnx",
                               providers=["CPUExecutionProvider"], provider_options=[{"device_id": 0}])

paddle_detection/detector.py

@@ -67,7 +67,10 @@ def request_book_areas(image):
         for img_str in img_str_list:
             img_data = base64.b64decode(img_str)
             np_array = np.frombuffer(img_data, np.uint8)
-            result.append(cv2.imdecode(np_array, cv2.IMREAD_COLOR))
+            img = cv2.imdecode(np_array, cv2.IMREAD_COLOR)
+            height, width = img.shape[:2]
+            if max(height, width) / min(height, width) <= 6.5:
+                result.append(img)  # 过滤异常结果
         return result
     else:
         return []