DEBUG

pp.py

@@ -1,5 +1,6 @@
 #!/usr/bin/env python3.10
 
+import gc
 import os
 import io
 import oss2
@@ -17,6 +18,7 @@ import paddle
 from PIL import Image, ImageFile
 from paddleocr import PaddleOCR
 
+paddle.set_flags({'FLAGS_fraction_of_gpu_memory_to_use': 0.6}) # 限制显存占用为GPU的80%
 logging.disable(logging.DEBUG)    # 关闭DEBUG日志的打印
 logging.disable(logging.WARNING)  # 关闭WARNING日志的打印
 warnings.filterwarnings("ignore")
@@ -84,6 +86,13 @@ JP = PaddleOCR(use_angle_cls=True, lang="japan")
 KR = PaddleOCR(use_angle_cls=True, lang="korean")
 RU = PaddleOCR(use_angle_cls=True, lang="ru")
 
+# 运行OCR并清理内存
+def process_ocr(model, image):
+    result = model.ocr(image, cls=True)[0] or []
+    paddle.device.cuda.empty_cache()  # 清理缓存
+    gc.collect()  # 强制垃圾回收
+    return result
+
 def process_images(conn, offset=0) -> int:
     with conn.cursor(pymysql.cursors.SSCursor) as cursor:
         cursor.execute("SELECT id, content FROM web_images WHERE text='' AND article_category_top_id=22 LIMIT 100 OFFSET %s", (offset,))
@@ -93,14 +102,20 @@ def process_images(conn, offset=0) -> int:
                 continue
             if isinstance(image, Image.Image):
                 image = np.array(image)
-            print('---------------------', id, content)
+            print(id, content)
 
             # 執行提取文字
-            en = EN.ocr(image, cls=True)[0] or []
-            ch = CH.ocr(image, cls=True)[0] or []
-            jp = JP.ocr(image, cls=True)[0] or []
-            kr = KR.ocr(image, cls=True)[0] or []
-            ru = RU.ocr(image, cls=True)[0] or []
+            #en = EN.ocr(image, cls=True)[0] or []
+            #ch = CH.ocr(image, cls=True)[0] or []
+            #jp = JP.ocr(image, cls=True)[0] or []
+            #kr = KR.ocr(image, cls=True)[0] or []
+            #ru = RU.ocr(image, cls=True)[0] or []
+            # 处理每个模型
+            ru = process_ocr(RU, image)
+            en = process_ocr(EN, image)
+            ch = process_ocr(CH, image)
+            jp = process_ocr(JP, image)
+            kr = process_ocr(KR, image)
 
             # 排除字符长度小于2的行, 排除纯数字的行, 排除置信度小于 0.8 的行
             jp = [x for x in jp if len(x[1][0]) > 1 and not x[1][0].isdigit() and x[1][1] > 0.8]
@@ -142,12 +157,17 @@ def process_images(conn, offset=0) -> int:
                 data = json.dumps(data, ensure_ascii=False, cls=MyEncoder)
                 c.execute("UPDATE web_images SET text = %s WHERE id = %s", (data, id))
                 conn.commit()
+            paddle.device.cuda.empty_cache()  # 清理缓存
+            gc.collect()  # 强制垃圾回收
+    paddle.device.cuda.empty_cache()  # 清理缓存
+    gc.collect()  # 强制垃圾回收
     return offset+100
 
 def main():
     conn = connect_to_mysql()
     offset = 2000
     while True:
+        print("LOOP:", offset)
         offset = process_images(conn, offset)
         time.sleep(0)