Update app.py

app.py

@@ -1,122 +1,94 @@
-# app.py — DeepSeek-OCR (HF Space, Claude Skill-ready)
-# - /ocr : REST API (POST)  → file / image_b64 / image_url 지원
-# - Gradio UI : 브라우저에서 업로드 테스트
-# 폴더 구조 전제:
-#   /app.py
-#   /DeepSeek-OCR-master/ (repo 그대로)
-#   /requirements.txt
-
-import io
-import os
-import sys
-import base64
-import traceback
-from typing import Optional
+# app.py — DeepSeek-OCR (HF Space, Claude Skill ready)
+# 지원: /ocr API (REST) + Gradio UI
+# 호환: DeepSeek-OCR-main / DeepSeek-OCR-master / DeepSeek-OCR-hf
 
+import io, os, sys, base64, traceback
+from typing import Optional
 from PIL import Image
 import numpy as np
-
 import gradio as gr
 from fastapi import FastAPI, UploadFile, File, Body
 from fastapi.responses import JSONResponse
 
 # ─────────────────────────────────────────────
-# 0) Repo 경로 추가
+# 1. 경로 자동 인식
 # ─────────────────────────────────────────────
 ROOT = os.path.dirname(__file__)
-DEEPSEEK_ROOT = os.path.join(ROOT, "DeepSeek-OCR-master")
-if DEEPSEEK_ROOT not in sys.path:
-    sys.path.append(DEEPSEEK_ROOT)
+CANDIDATES = [
+    "DeepSeek-OCR-master",
+    "DeepSeek-OCR-hf",
+    os.path.join("DeepSeek-OCR-main", "DeepSeek-OCR-master"),
+    os.path.join("DeepSeek-OCR-main", "DeepSeek-OCR-hf"),
+]
+for rel in CANDIDATES:
+    absdir = os.path.join(ROOT, rel)
+    if os.path.isdir(absdir) and absdir not in sys.path:
+        sys.path.append(absdir)
+        print(f"[path] added: {absdir}")
 
 # ─────────────────────────────────────────────
-# 1) DeepSeek-OCR 어댑터
-#   - 저장소가 제공하는 실제 진입점 이름이 다를 수 있어
-#     여러 패턴을 시도하도록 구성했습니다.
-#   - 필요 시 아래 "TODO" 부분을 실제 함수명으로 바꾸세요.
+# 2. DeepSeek-OCR 어댑터
 # ─────────────────────────────────────────────
 class DeepSeekOCRAdapter:
     def __init__(self):
-        """
-        가능한 엔트리 시나리오:
-        A) deeps eek_ocr.py 내부에 클래스/함수 제공
-           - class DeepSeekOCR  → .recognize(Image) 반환
-           - def ocr_image(Image, lang="auto") 반환
-        B) run_dpsk_ocr_image.py 내부에 함수 제공
-           - def infer(Image) 또는 def run(Image, ...) 등
-        """
-        self.backend = None
-        self.fn = None  # callable(image, lang='auto') -> str
-
-        # A-1) class DeepSeekOCR 시도
+        self.fn = None
+
+        # (1) deepseek_ocr.py
         try:
-            import deepseek_ocr as dso  # DeepSeek-OCR-master/deeps eek_ocr.py
+            import deepseek_ocr as dso
+            if hasattr(dso, "ocr_image"):
+                self.fn = lambda img, lang="auto": dso.ocr_image(img, lang=lang)
+                print("[Adapter] Using deepseek_ocr.ocr_image()")
+                return
             if hasattr(dso, "DeepSeekOCR"):
-                self.backend = dso.DeepSeekOCR()
-                def _call(image: Image.Image, lang="auto"):
-                    # 클래스가 recognize(image, lang) 보유한다고 가정
-                    if hasattr(self.backend, "recognize"):
-                        return self.backend.recognize(image, lang=lang)
-                    # 혹은 run/image_to_text 등의 이름일 수 있음
-                    for cand in ("run", "infer", "image_to_text", "predict"):
-                        if hasattr(self.backend, cand):
-                            return getattr(self.backend, cand)(image)
-                    raise AttributeError("DeepSeekOCR class found but no callable method.")
-                self.fn = _call
-                print("[DeepSeekOCRAdapter] Using deeps eek_ocr.DeepSeekOCR")
+                model = dso.DeepSeekOCR()
+                self.fn = lambda img, lang="auto": model.recognize(img, lang=lang)
+                print("[Adapter] Using deepseek_ocr.DeepSeekOCR()")
                 return
         except Exception as e:
-            print("[DeepSeekOCRAdapter] A-1 fallback:", e)
+            print("[Adapter] deepseek_ocr import failed:", e)
 
-        # A-2) 함수형 ocr_image 시도
+        # (2) run_dpsk_ocr.py (HF용)
         try:
-            import deepseek_ocr as dso
-            if hasattr(dso, "ocr_image"):
-                def _call(image: Image.Image, lang="auto"):
-                    return dso.ocr_image(image, lang=lang)  # TODO: 필요 시 인자명 맞추기
-                self.fn = _call
-                print("[DeepSeekOCRAdapter] Using deeps eek_ocr.ocr_image")
+            import run_dpsk_ocr as runner
+            if hasattr(runner, "run"):
+                self.fn = lambda img, lang="auto": runner.run(img)
+                print("[Adapter] Using run_dpsk_ocr.run()")
+                return
+            if hasattr(runner, "infer"):
+                self.fn = lambda img, lang="auto": runner.infer(img)
+                print("[Adapter] Using run_dpsk_ocr.infer()")
                 return
         except Exception as e:
-            print("[DeepSeekOCRAdapter] A-2 fallback:", e)
+            print("[Adapter] run_dpsk_ocr import failed:", e)
 
-        # B) run_dpsk_ocr_image.py 스크립트형 시도
+        # (3) run_dpsk_ocr_image.py (CLI 스타일)
         try:
             import run_dpsk_ocr_image as runner
-            for cand in ("infer", "run", "predict", "main"):
-                if hasattr(runner, cand) and callable(getattr(runner, cand)):
-                    def _call(image: Image.Image, lang="auto", _fn=getattr(runner, cand)):
-                        # NOTE: 해당 함수가 PIL.Image가 아닌 파일경로를 요구할 수 있습니다.
-                        # 그런 경우 임시 파일로 저장해 넘깁니다.
-                        try:
-                            return _fn(image)  # PIL.Image 직접 받는 케이스
-                        except Exception:
-                            import tempfile
-                            with tempfile.NamedTemporaryFile(suffix=".png", delete=True) as tmp:
-                                image.save(tmp.name)
-                                # 가장 흔한 CLI 스타일: (path)만 or (path, config)
-                                try:
-                                    return _fn(tmp.name)
-                                except Exception:
-                                    # 반환이 dict/text 등 무엇이든 str로 캐스팅
-                                    return str(_fn(tmp.name))
+            for cand in ("run", "infer", "main"):
+                if hasattr(runner, cand):
+                    fn = getattr(runner, cand)
+                    def _call(img, lang="auto", _fn=fn):
+                        import tempfile
+                        with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp:
+                            img.save(tmp.name)
+                            return str(_fn(tmp.name))
                     self.fn = _call
-                    print(f"[DeepSeekOCRAdapter] Using run_dpsk_ocr_image.{cand}")
+                    print(f"[Adapter] Using run_dpsk_ocr_image.{cand}()")
                     return
         except Exception as e:
-            print("[DeepSeekOCRAdapter] B fallback:", e)
+            print("[Adapter] run_dpsk_ocr_image import failed:", e)
 
-        # 마지막 안전장치: 데모
-        print("[DeepSeekOCRAdapter] No concrete entry found. Falling back to DEMO.")
-        def _demo(image: Image.Image, lang="auto"):
-            return "[DEMO] 연결 완료 — 실제 함수명을 app.py에서 한 줄만 바꿔주세요."
-        self.fn = _demo
+        # fallback
+        self.fn = lambda img, lang="auto": "[DEMO] 연결 성공 — 실제 추론 함수 확인 필요."
+        print("[Adapter] ⚠️ DEMO fallback active.")
 
-    def recognize(self, image: Image.Image, lang: str = "auto") -> str:
-        return self.fn(image, lang=lang)
+    def recognize(self, image: Image.Image, lang="auto"):
+        return self.fn(image, lang)
 
 
 # ─────────────────────────────────────────────
-# 2) 유틸
+# 3. 유틸
 # ─────────────────────────────────────────────
 def _to_pil(x) -> Image.Image:
     if isinstance(x, Image.Image):
@@ -128,8 +100,7 @@ def _to_pil(x) -> Image.Image:
     raise TypeError("Unsupported image type")
 
 def _b64_to_image(image_b64: str) -> Image.Image:
-    raw = base64.b64decode(image_b64)
-    return _to_pil(raw)
+    return _to_pil(base64.b64decode(image_b64))
 
 def _url_to_image(url: str) -> Image.Image:
     import requests
@@ -139,55 +110,49 @@ def _url_to_image(url: str) -> Image.Image:
 
 
 # ─────────────────────────────────────────────
-# 3) FastAPI (REST)
+# 4. FastAPI
 # ─────────────────────────────────────────────
 api = FastAPI(title="DeepSeek-OCR API")
 _engine = DeepSeekOCRAdapter()
 
 @api.post("/ocr")
-async def ocr(
+async def ocr_endpoint(
     image_b64: Optional[str] = Body(default=None),
     image_url: Optional[str] = Body(default=None),
     lang: str = Body(default="auto"),
     file: Optional[UploadFile] = File(default=None),
 ):
     try:
-        if file is not None:
+        if file:
             image = _to_pil(await file.read())
         elif image_b64:
             image = _b64_to_image(image_b64)
         elif image_url:
             image = _url_to_image(image_url)
         else:
-            return JSONResponse(status_code=400, content={
-                "ok": False, "error": "Provide one of: file | image_b64 | image_url"
-            })
-        text = _engine.recognize(image, lang=lang)
+            return JSONResponse(status_code=400, content={"ok": False, "error": "No image input"})
+        text = _engine.recognize(image, lang)
         return {"ok": True, "text": text}
     except Exception as e:
-        return JSONResponse(status_code=500, content={
-            "ok": False, "error": str(e), "trace": traceback.format_exc()
-        })
+        return JSONResponse(status_code=500, content={"ok": False, "error": str(e), "trace": traceback.format_exc()})
+
 
 # ─────────────────────────────────────────────
-# 4) Gradio UI (테스트)
+# 5. Gradio UI
 # ─────────────────────────────────────────────
-def _predict(image, lang):
-    if image is None:
-        return "No image."
-    pil = _to_pil(image)
-    return _engine.recognize(pil, lang=lang)
+def gradio_predict(img, lang):
+    if img is None:
+        return "No image provided."
+    return _engine.recognize(_to_pil(img), lang)
 
-with gr.Blocks(title="DeepSeek-OCR (Claude-ready)") as demo:
+with gr.Blocks(title="DeepSeek-OCR (Claude Ready)") as demo:
     gr.Markdown("### DeepSeek-OCR (HF Space)\n이미지를 업로드하면 텍스트를 추출합니다.")
     with gr.Row():
-        img = gr.Image(type="pil", label="Input image")
+        img = gr.Image(type="pil", label="Input Image")
         out = gr.Textbox(label="OCR Result", lines=8)
     lang = gr.Radio(["auto","en","ko","ja","zh"], value="auto", label="Language")
     btn = gr.Button("Run OCR")
-    btn.click(_predict, inputs=[img, lang], outputs=[out])
+    btn.click(gradio_predict, inputs=[img, lang], outputs=[out])
 
-# HF Spaces는 보통 Gradio 앱을 기본 엔트리로 띄우지만,
-# FastAPI 엔드포인트도 함께 노출하려면 아래처럼 alias를 둡니다.
-app = api
-demo.queue(concurrency_count=1)
+app = api  # FastAPI 엔진 노출
+demo.queue()  # 최신 gradio 버전에 맞게 수정