统一模型接口，新增文本分类接口

docker-compose.yml

@@ -20,13 +20,13 @@ x-paddle:
     - ./tmp_img:/app/tmp_img
 
 services:
-  ocr_api:
+  ocr:
     <<: *paddle_template
     build:
       context: ./services/paddle_services
-    container_name: ocr_api
+    container_name: ocr
-    hostname: ocr_api
+    hostname: ocr
-    command: [ '-w', '1', 'ocr_api:app', '--bind', '0.0.0.0:5001' ]
+    command: [ '-w', '1', 'ocr:app', '--bind', '0.0.0.0:5001' ]
     deploy:
       resources:
         reservations:
@@ -35,11 +35,11 @@ services:
               capabilities: [ 'gpu' ]
               driver: 'nvidia'
 
-  settlement_api:
+  ie_settlement:
     <<: *paddle_template
-    container_name: settlement_api
+    container_name: ie_settlement
-    hostname: settlement_api
+    hostname: ie_settlement
-    command: [ '-w', '1', 'settlement_api:app', '--bind', '0.0.0.0:5002' ]
+    command: [ '-w', '1', 'ie_settlement:app', '--bind', '0.0.0.0:5002' ]
     deploy:
       resources:
         reservations:
@@ -48,11 +48,11 @@ services:
               capabilities: [ 'gpu' ]
               driver: 'nvidia'
 
-  discharge_api:
+  ie_discharge:
     <<: *paddle_template
-    container_name: discharge_api
+    container_name: ie_discharge
-    hostname: discharge_api
+    hostname: ie_discharge
-    command: [ '-w', '1', 'discharge_api:app', '--bind', '0.0.0.0:5003' ]
+    command: [ '-w', '1', 'ie_discharge:app', '--bind', '0.0.0.0:5003' ]
     deploy:
       resources:
         reservations:
@@ -61,11 +61,11 @@ services:
               capabilities: [ 'gpu' ]
               driver: 'nvidia'
 
-  cost_api:
+  ie_cost:
     <<: *paddle_template
-    container_name: cost_api
+    container_name: ie_cost
-    hostname: cost_api
+    hostname: ie_cost
-    command: [ '-w', '1', 'cost_api:app', '--bind', '0.0.0.0:5004' ]
+    command: [ '-w', '1', 'ie_cost:app', '--bind', '0.0.0.0:5004' ]
     deploy:
       resources:
         reservations:
@@ -74,11 +74,11 @@ services:
               capabilities: [ 'gpu' ]
               driver: 'nvidia'
 
-  clas_api:
+  clas_orientation:
     <<: *paddle_template
-    container_name: clas_api
+    container_name: clas_orientation
-    hostname: clas_api
+    hostname: clas_orientation
-    command: [ '-w', '1', 'clas_api:app', '--bind', '0.0.0.0:5005' ]
+    command: [ '-w', '1', 'clas_orientation:app', '--bind', '0.0.0.0:5005' ]
     deploy:
       resources:
         reservations:
@@ -87,11 +87,11 @@ services:
               capabilities: [ 'gpu' ]
               driver: 'nvidia'
 
-  det_api:
+  det_book:
     <<: *paddle_template
-    container_name: det_api
+    container_name: det_book
-    hostname: det_api
+    hostname: det_book
-    command: [ '-w', '1', 'det_api:app', '--bind', '0.0.0.0:5006' ]
+    command: [ '-w', '1', 'det_book:app', '--bind', '0.0.0.0:5006' ]
     deploy:
       resources:
         reservations:
@@ -100,11 +100,11 @@ services:
               capabilities: [ 'gpu' ]
               driver: 'nvidia'
 
-  dewarp_api:
+  dewarp:
     <<: *paddle_template
-    container_name: dewarp_api
+    container_name: dewarp
-    hostname: dewarp_api
+    hostname: dewarp
-    command: [ '-w', '1', 'dewarp_api:app', '--bind', '0.0.0.0:5007' ]
+    command: [ '-w', '1', 'dewarp:app', '--bind', '0.0.0.0:5007' ]
     deploy:
       resources:
         reservations:
@@ -129,37 +129,37 @@ services:
 #      - dewarp_api
 #    command: [ 'photo_review.py', '--clean', 'True' ]
 
-  #  photo_review_2:
+#  photo_review_2:
-  #    <<: *project_template
+#    <<: *project_template
-  #    container_name: photo_review_2
+#    container_name: photo_review_2
-  #    hostname: photo_review_2
+#    hostname: photo_review_2
-  #    depends_on:
+#    depends_on:
-  #      - photo_review_1
+#      - photo_review_1
-  #    command: [ 'photo_review.py' ]
+#    command: [ 'photo_review.py' ]
-  #
+#
-  #  photo_review_3:
+#  photo_review_3:
-  #    <<: *project_template
+#    <<: *project_template
-  #    container_name: photo_review_3
+#    container_name: photo_review_3
-  #    hostname: photo_review_3
+#    hostname: photo_review_3
-  #    depends_on:
+#    depends_on:
-  #      - photo_review_2
+#      - photo_review_2
-  #    command: [ 'photo_review.py' ]
+#    command: [ 'photo_review.py' ]
-  #
+#
-  #  photo_review_4:
+#  photo_review_4:
-  #    <<: *project_template
+#    <<: *project_template
-  #    container_name: photo_review_4
+#    container_name: photo_review_4
-  #    hostname: photo_review_4
+#    hostname: photo_review_4
-  #    depends_on:
+#    depends_on:
-  #      - photo_review_3
+#      - photo_review_3
-  #    command: [ 'photo_review.py' ]
+#    command: [ 'photo_review.py' ]
-  #
+#
-  #  photo_review_5:
+#  photo_review_5:
-  #    <<: *project_template
+#    <<: *project_template
-  #    container_name: photo_review_5
+#    container_name: photo_review_5
-  #    hostname: photo_review_5
+#    hostname: photo_review_5
-  #    depends_on:
+#    depends_on:
-  #      - photo_review_4
+#      - photo_review_4
-  #    command: [ 'photo_review.py' ]
+#    command: [ 'photo_review.py' ]
 
 #  photo_mask_1:
 #    <<: *project_template

photo_review/auto_photo_review.py

@@ -185,8 +185,8 @@ def information_extraction(ie, phrecs, identity):
             target_images = model_util.request_book_areas(img_path)  # 识别文档区域并裁剪
             angle_count = defaultdict(int, {'0': 0})  # 分割后图片的最优角度统计
             for target_image in target_images:
-                dewarped_image = model_util.request_dewarped_image(target_image)  # 去扭曲
+                dewarped_image = model_util.dewarp(target_image)  # 去扭曲
-                angles = model_util.request_image_orientation(dewarped_image)
+                angles = model_util.clas_orientation(dewarped_image)
 
                 split_results = image_util.split(dewarped_image)
                 for split_result in split_results:

services/paddle_services/clas_orientation.py

@@ -1,4 +1,4 @@
-import logging
+import logging.config
 
 from flask import Flask, request
 from paddleclas import PaddleClas
@@ -10,9 +10,9 @@ app = Flask(__name__)
 CLAS = PaddleClas(model_name='text_image_orientation')
 
 
-@app.route('/clas/orientation', methods=['POST'])
+@app.route(rule='/', methods=['POST'])
 @process_request
-def orientation():
+def main():
     """
     判断图片旋转角度，逆时针旋转该角度后为正。可能值['0', '90', '180', '270']
     :return: 最有可能的两个角度

services/paddle_services/clas_text.py

@@ -0,0 +1,28 @@
+import logging.config
+
+from flask import Flask, request
+from paddlenlp import Taskflow
+
+from log import LOGGING_CONFIG
+from utils import process_request
+
+app = Flask(__name__)
+schema = ['基本医保结算单', '出院记录', '费用清单']
+CLAS = Taskflow('zero_shot_text_classification', model='utc-xbase', schema=schema,
+                task_path='model/text_classification', precision='fp16')
+
+
+@app.route('/', methods=['POST'])
+@process_request
+def main():
+    text = request.form.get('text')
+    cls_result = CLAS(text)
+    cls_result = cls_result[0].get('predictions')[0]
+    if cls_result['score'] < 0.8:
+        raise Exception(f'识别结果置信度过低！text: {text}')
+    return cls_result['label']
+
+
+if __name__ == '__main__':
+    logging.config.dictConfig(LOGGING_CONFIG)
+    app.run('0.0.0.0', 5008)

services/paddle_services/det_book.py

@@ -1,4 +1,4 @@
-import logging
+import logging.config
 import os.path
 
 import cv2
@@ -11,9 +11,9 @@ from utils import process_request, parse_img_path
 app = Flask(__name__)
 
 
-@app.route('/det/books', methods=['POST'])
+@app.route('/', methods=['POST'])
 @process_request
-def books():
+def main():
     img_path = request.form.get('img_path')
     result = detector.get_book_areas(img_path)
 

services/paddle_services/dewarp.py

@@ -1,4 +1,4 @@
-import logging
+import logging.config
 import os
 
 import cv2
@@ -11,9 +11,9 @@ from utils import process_request, parse_img_path
 app = Flask(__name__)
 
 
-@app.route('/dewarp', methods=['POST'])
+@app.route('/', methods=['POST'])
 @process_request
-def dewarp():
+def main():
     img_path = request.form.get('img_path')
     img = cv2.imread(img_path)
     dewarped_img = dewarper.dewarp_image(img)

services/paddle_services/ie_cost.py

@@ -1,5 +1,5 @@
 import json
-import logging
+import logging.config
 
 from flask import Flask, request
 from paddlenlp import Taskflow
@@ -14,9 +14,9 @@ COST = Taskflow('information_extraction', schema=COST_LIST_SCHEMA, model='uie-x-
                 task_path='model/cost_list_model', layout_analysis=False, precision='fp16')
 
 
-@app.route('/nlp/cost', methods=['POST'])
+@app.route('/', methods=['POST'])
 @process_request
-def cost():
+def main():
     img_path = request.form.get('img_path')
     layout = request.form.get('layout')
     return COST({'doc': img_path, 'layout': json.loads(layout)})

services/paddle_services/ie_discharge.py

@@ -1,5 +1,5 @@
 import json
-import logging
+import logging.config
 
 from flask import Flask, request
 from paddlenlp import Taskflow
@@ -16,9 +16,9 @@ DISCHARGE = Taskflow('information_extraction', schema=DISCHARGE_RECORD_SCHEMA, m
                      task_path='model/discharge_record_model', layout_analysis=False, precision='fp16')
 
 
-@app.route('/nlp/discharge', methods=['POST'])
+@app.route('/', methods=['POST'])
 @process_request
-def discharge():
+def main():
     img_path = request.form.get('img_path')
     layout = request.form.get('layout')
     return DISCHARGE({'doc': img_path, 'layout': json.loads(layout)})

services/paddle_services/ie_settlement.py

@@ -1,5 +1,5 @@
 import json
-import logging
+import logging.config
 
 from flask import Flask, request
 from paddlenlp import Taskflow
@@ -20,9 +20,9 @@ SETTLEMENT_IE = Taskflow('information_extraction', schema=SETTLEMENT_LIST_SCHEMA
                          task_path='model/settlement_list_model', layout_analysis=False, precision='fp16')
 
 
-@app.route('/nlp/settlement', methods=['POST'])
+@app.route('/', methods=['POST'])
 @process_request
-def settlement():
+def main():
     img_path = request.form.get('img_path')
     layout = request.form.get('layout')
     return SETTLEMENT_IE({'doc': img_path, 'layout': json.loads(layout)})

services/paddle_services/model/text_classification/README.md

@@ -0,0 +1 @@
+文本分类模型存放目录

services/paddle_services/ocr.py

@@ -1,4 +1,4 @@
-import logging
+import logging.config
 
 from flask import Flask, request
 from paddleocr import PaddleOCR
@@ -10,9 +10,9 @@ app = Flask(__name__)
 OCR = PaddleOCR(use_angle_cls=False, show_log=False, gpu_id=0, det_db_box_thresh=0.3)
 
 
-@app.route('/ocr', methods=['POST'])
+@app.route('/', methods=['POST'])
 @process_request
-def ocr():
+def main():
     img_path = request.form.get('img_path')
     return OCR.ocr(img_path, cls=False)
 

util/common_util.py

@@ -36,7 +36,7 @@ def get_ocr_layout(ocr, img_path):
         return True
 
     layout = []
-    ocr_result = model_util.request_ocr(img_path)
+    ocr_result = model_util.ocr(img_path)
     ocr_result = ocr_result[0]
     if not ocr_result:
         return layout

util/model_util.py

@@ -7,13 +7,13 @@ from tenacity import retry, stop_after_attempt, wait_random
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('OCR识别失败！'))
-def request_ocr(img_path):
+def ocr(img_path):
     """
     请求图片OCR识别接口
     :param img_path: 待识别图片路径
     :return: 识别结果
     """
-    url = 'http://ocr_api:5001/ocr'
+    url = 'http://ocr:5001'
     response = requests.post(url, {'img_path': img_path})
     if response.status_code == 200:
         return response.json()
@@ -23,14 +23,14 @@ def request_ocr(img_path):
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('抽取基本医保结算单失败！'))
-def request_discharge_info(img_path, layout):
+def ie_settlement(img_path, layout):
     """
     请求基本医保结算单信息抽取接口
     :param img_path: 待抽取图片路径
     :param layout: 图片ocr信息
     :return: 抽取结果
     """
-    url = 'http://settlement_api:5002/nlp/settlement'
+    url = 'http://ie_settlement:5002'
     response = requests.post(url, {'img_path': img_path, 'layout': json.dumps(layout)})
     if response.status_code == 200:
         return response.json()
@@ -40,14 +40,14 @@ def request_discharge_info(img_path, layout):
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('抽取出院记录失败！'))
-def request_discharge_info(img_path, layout):
+def ie_discharge(img_path, layout):
     """
     请求出院记录信息抽取接口
     :param img_path: 待抽取图片路径
     :param layout: 图片ocr信息
     :return: 抽取结果
     """
-    url = 'http://discharge_api:5003/nlp/discharge'
+    url = 'http://ie_discharge:5003'
     response = requests.post(url, {'img_path': img_path, 'layout': json.dumps(layout)})
     if response.status_code == 200:
         return response.json()
@@ -57,14 +57,14 @@ def request_discharge_info(img_path, layout):
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('抽取费用清单失败！'))
-def request_cost_info(img_path, layout):
+def ie_cost(img_path, layout):
     """
     请求费用清单信息抽取接口
     :param img_path: 待抽取图片路径
     :param layout: 图片ocr信息
     :return: 抽取结果
     """
-    url = 'http://cost_api:5004/nlp/cost'
+    url = 'http://ie_cost:5004'
     response = requests.post(url, {'img_path': img_path, 'layout': json.dumps(layout)})
     if response.status_code == 200:
         return response.json()
@@ -74,13 +74,13 @@ def request_cost_info(img_path, layout):
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('获取图片方向失败！'))
-def request_image_orientation(img_path):
+def clas_orientation(img_path):
     """
     请求图片方向分类接口
     :param img_path: 待分类图片路径
     :return: 最有可能的两个图片方向
     """
-    url = 'http://clas_api:5005/clas/orientation'
+    url = 'http://clas_orientation:5005'
     response = requests.post(url, {'img_path': img_path})
     if response.status_code == 200:
         return response.json()
@@ -90,13 +90,13 @@ def request_image_orientation(img_path):
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('获取文档区域失败！'))
-def request_book_areas(img_path):
+def det_book(img_path):
     """
     请求文档区域识别接口
     :param img_path: 待识别图片路径
     :return: 文档图片路径列表
     """
-    url = 'http://det_api:5006/det/books'
+    url = 'http://det_book:5006'
     response = requests.post(url, {'img_path': img_path})
     if response.status_code == 200:
         return response.json()
@@ -106,15 +106,33 @@ def request_book_areas(img_path):
 
 @retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
        after=lambda x: logging.warning('矫正扭曲图片失败！'))
-def request_dewarped_image(img_path):
+def dewarp(img_path):
     """
     请求矫正图片接口
     :param img_path: 待矫正图片路径
     :return: 矫正后的图片路径
     """
-    url = 'http://dewarp_api:5007/dewarp'
+    url = 'http://dewarp:5007'
     response = requests.post(url, {'img_path': img_path})
     if response.status_code == 200:
         return response.json()
     else:
         return img_path
+
+
+@retry(stop=stop_after_attempt(3), wait=wait_random(1, 3), reraise=True,
+       after=lambda x: logging.warning('文本分类失败！'))
+def clas_text(text):
+    """
+    请求文本分类接口
+    :param text: 待分类文本
+    :return: 分类结果
+    """
+    if not text:
+        return None
+    url = 'http://clas_text:5008'
+    response = requests.post(url, {'text': text})
+    if response.status_code == 200:
+        return response.json()
+    else:
+        return None