Update app.py

app.py

@@ -1,30 +1,22 @@
 import atexit
 import functools
+import os
+from pathlib import Path
 from queue import Queue
 from threading import Event, Thread
 
-from paddleocr import PaddleOCR, draw_ocr
-from PIL import Image
+from paddleocr import PaddleOCR
 import gradio as gr
-import numpy as np
-import io
-
 
 LANG_CONFIG = {
     "ch": {"num_workers": 1},
     "en": {"num_workers": 1},
-    # "fr": {"num_workers": 1},
-    # "german": {"num_workers": 1},
-    # "korean": {"num_workers": 1},
-    # "japan": {"num_workers": 1},
 }
 CONCURRENCY_LIMIT = 8
-
+TMP_PREFIX = "/tmp/gradio/"
 
 class PaddleOCRModelManager(object):
-    def __init__(self,
-                 num_workers,
-                 model_factory):
+    def __init__(self, num_workers, model_factory):
         super().__init__()
         self._model_factory = model_factory
         self._queue = Queue()
@@ -38,7 +30,6 @@ class PaddleOCRModelManager(object):
             self._workers.append(worker)
 
     def infer(self, *args, **kwargs):
-        # XXX: Should I use a more lightweight data structure, say, a future?
         result_queue = Queue(maxsize=1)
         self._queue.put((args, kwargs, result_queue))
         success, payload = result_queue.get()
@@ -69,67 +60,70 @@ class PaddleOCRModelManager(object):
             finally:
                 self._queue.task_done()
 
-
 def create_model(lang):
     return PaddleOCR(lang=lang, use_angle_cls=True, use_gpu=False)
 
-
 model_managers = {}
 for lang, config in LANG_CONFIG.items():
-    model_manager = PaddleOCRModelManager(config["num_workers"], functools.partial(create_model, lang=lang))
+    model_manager = PaddleOCRModelManager(
+        config["num_workers"], functools.partial(create_model, lang=lang)
+    )
     model_managers[lang] = model_manager
 
-
 def close_model_managers():
     for manager in model_managers.values():
         manager.close()
 
-
-# XXX: Not sure if gradio allows adding custom teardown logic
 atexit.register(close_model_managers)
 
-
-def inference(img, lang):
-    # 将各种可能类型统一转为 np.ndarray
-    if img is None:
-        raise ValueError("输入图片为空")
-
-    if isinstance(img, Image.Image):
-        img_np = np.array(img.convert("RGB"))
-    elif isinstance(img, np.ndarray):
-        img_np = img
-    elif isinstance(img, bytes):
-        img_np = np.array(Image.open(io.BytesIO(img)).convert("RGB"))
-    elif isinstance(img, str):
-        # 兼容仍然传入文件路径的情况
-        img_np = np.array(Image.open(img).convert("RGB"))
-    else:
-        raise TypeError(f"不支持的图片类型: {type(img)}")
+def _safe_cleanup(path_str: str):
+    # 仅清理 /tmp/gradio 下的文件与空目录，避免误删
+    try:
+        real = Path(os.path.realpath(path_str))
+        if str(real).startswith(TMP_PREFIX) and real.is_file():
+            real.unlink(missing_ok=True)
+            parent = real.parent
+            # 尝试删除空目录（/tmp/gradio/<hash>）
+            if str(parent).startswith(TMP_PREFIX):
+                try:
+                    parent.rmdir()
+                except OSError:
+                    pass
+    except Exception:
+        # 清理失败不影响主流程
+        pass
+
+def inference(img_path: str, lang: str):
+    # Gradio Image(type='filepath') 传入的是服务端本地路径字符串
+    if not isinstance(img_path, str) or not img_path:
+        raise ValueError("无效的图片路径")
 
     ocr = model_managers[lang]
-    result = ocr.infer(img_np, cls=True)[0]
-    return result
+    # 直接使用路径字符串，PaddleOCR 支持 str
+    result = ocr.infer(img_path, cls=True)[0]
 
+    # 推理完成后清理临时文件
+    _safe_cleanup(img_path)
+    return result
 
-title = 'PaddleOCR'
-description = '''
-- Gradio demo for PaddleOCR. PaddleOCR demo supports Chinese, English, French, German, Korean and Japanese. 
-- To use it, simply upload your image and choose a language from the dropdown menu, or click one of the examples to load them. Read more at the links below.
-- [Docs](https://paddlepaddle.github.io/PaddleOCR/), [Github Repository](https://github.com/PaddlePaddle/PaddleOCR).
-'''
+title = "PaddleOCR"
+description = """
+- Gradio demo for PaddleOCR. Supports Chinese and English.
+- Upload an image and choose language; returns structured JSON.
+"""
 
 examples = [
-    ['en_example.jpg','en'],
-    ['cn_example.jpg','ch'],
-    ['jp_example.jpg','japan'],
+    ["en_example.jpg", "en"],
+    ["cn_example.jpg", "ch"],
 ]
 
 css = ".output_image, .input_image {height: 40rem !important; width: 100% !important;}"
+
 gr.Interface(
     inference,
     [
-        gr.Image(type='pil', label='Input'),
-        gr.Dropdown(choices=list(LANG_CONFIG.keys()), value='en', label='language')
+        gr.Image(type="filepath", label="Input"),
+        gr.Dropdown(choices=list(LANG_CONFIG.keys()), value="en", label="language"),
     ],
     gr.JSON(label="Output"),
     title=title,
@@ -138,4 +132,4 @@ gr.Interface(
     cache_examples=False,
     css=css,
     concurrency_limit=CONCURRENCY_LIMIT,
-    ).launch(debug=False)
+).launch(debug=False)