Upload 2 files

app.py

@@ -1,20 +1,15 @@
 import os
-import tempfile
-import time
-import logging
-import base64
-from typing import Optional
-
+import gradio as gr
 import fitz  # PyMuPDF
+from pathlib import Path
 import google.generativeai as genai
-from pydantic import BaseModel, Field, HttpUrl
-from fastapi import FastAPI, HTTPException, BackgroundTasks
-from fastapi.responses import JSONResponse
+import tempfile
+import base64
 from concurrent.futures import ThreadPoolExecutor, as_completed
-import httpx  # HTTPクライアントライブラリ
-
+import logging
+import time
 
-# ロギング設定 (FastAPI のログレベルに合わせる)
+# ロギング設定
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
 # 環境変数から設定を読み込む
@@ -28,7 +23,7 @@ if not MODEL_NAME:
 
 genai.configure(api_key=GOOGLE_API_KEY)
 
-# --- PDF 処理関数 (変更なし) ---
+
 def split_pdf(pdf_path, output_dir, pages_per_chunk=5):
     """PDFを指定ページ数ごとに分割する関数"""
     pdf_document = fitz.open(pdf_path)
@@ -38,12 +33,19 @@ def split_pdf(pdf_path, output_dir, pages_per_chunk=5):
 
     for start_page in range(0, total_pages, pages_per_chunk):
         end_page = min(start_page + pages_per_chunk - 1, total_pages - 1)
+
+        # 新しいPDFドキュメントを作成
         output_pdf = fitz.open()
+
+        # 指定範囲のページを新しいPDFに追加
         for page_num in range(start_page, end_page + 1):
             output_pdf.insert_pdf(pdf_document, from_page=page_num, to_page=page_num)
+
+        # 分割したPDFを保存
         output_path = os.path.join(output_dir, f"split_{start_page+1}_to_{end_page+1}.pdf")
         output_pdf.save(output_path)
         output_pdf.close()
+
         split_pdfs.append((start_page, output_path))
 
     pdf_document.close()
@@ -66,12 +68,48 @@ def ocr_pdf_with_gemini(pdf_path, model_name):
 
     **Crucial Instructions:**
 
-    1.  **Consistency:** Maintain a consistent Markdown style (headings, lists, tables, etc.)
-    2.  **Visual Fidelity:** Strive to reproduce the original visual appearance.
-    3.  **OCR Correction:** Correct OCR errors (typos, misspellings, etc.).
-    4.  **Content Filtering (Screenshots):** Focus *exclusively* on the *main content area*.  *Do not* include text from window title bars, OS toolbars.
-    5. **Accuracy and Clarity:** Prioritize providing accurate and clear information.
-    6. **Output:** Only provide the extracted text in Markdown. No extra chat, greetings.
+    1.  **Consistency:** Maintain a consistent Markdown style (headings, lists, tables, font sizes, etc.) throughout the entire output, even if the PDF is split into multiple parts. Avoid variations in formatting between different sections.
+    2.  **Visual Fidelity:** Strive to reproduce the original visual appearance of the PDF as closely as possible in Markdown. Pay close attention to:
+        *   **Tables:** Recreate tables with accurate borders, cell alignment, and row/column spans.  Use extended Markdown table syntax if needed.
+        *   **Lists:** Preserve the original list style (numbered, bulleted, nested).
+        *   **Font Sizes:**  Use Markdown heading levels (#, ##, ###) and relative font size adjustments (if possible in your Markdown flavor) to approximate the visual hierarchy of the original text.  Larger text should generally correspond to higher-level headings.
+        *   **Layout:** Try to maintain the spatial relationships between elements (e.g., paragraphs, images). If there are multiple columns, consider representing that structure in Markdown, perhaps using tables.
+    3.  **OCR Correction:** The OCR process may introduce errors (typos, misspellings, incorrect characters).  Apply your language understanding capabilities to correct these errors and produce grammatically correct and semantically meaningful text.  Do *not* blindly output the raw OCR result if it contains obvious mistakes.
+    4.  **Content Filtering (Screenshots):** If the PDF primarily contains screenshots (e.g., of software interfaces), focus *exclusively* on extracting text from the *main content area* of the screenshots.  *Do not* include text from:
+        *   Window title bars
+        *   Operating system toolbars (e.g., Windows taskbar)
+        *   Menu bars *unless* they are directly related to the primary content (e.g. describing a software's menu options)
+        *   Any other UI elements that are not part of the core content being displayed.
+    5. **Accuracy and Clarity:** Prioritize providing accurate and clear information to the user. Do not simply reproduce OCR output verbatim if it is nonsensical or misleading. Use your understanding of the content to present information in a user-friendly way.
+    6. **Output:** Only provide the extracted text in Markdown.
+
+    **Example (Illustrative - Adapt to the specific PDF):**
+
+    **Input PDF (Screenshot of a webpage):**
+
+    ```
+    [Screenshot of a webpage with a large heading "Welcome", a paragraph of text, a bulleted list, and a table.]
+    ```
+
+    **Desired Markdown Output:**
+
+    ```markdown
+    # Welcome
+
+    This is a paragraph of introductory text. It explains the purpose of the webpage and provides some context.
+
+    *   This is the first bullet point.
+    *   This is the second bullet point.
+        *   This is a nested bullet point.
+
+    | Feature      | Description                | Price   |
+    |--------------|----------------------------|---------|
+    | Feature A    | Description of Feature A  | $10     |
+    | Feature B    | Description of Feature B  | $20     |
+
+    ```
+
+    **Do NOT include text like "File Edit View" (from a menu bar) or "[X] Minimize Maximize Close" (from a window title bar).**
     """
 
     try:
@@ -92,134 +130,108 @@ def ocr_pdf_with_gemini(pdf_path, model_name):
 
 
 
-def process_pdf(pdf_path, model_name):
-    """PDFファイルを処理するメイン関数 (同期処理に変更)"""
+def process_pdf(pdf_file, progress=gr.Progress()):
+    """PDFファイルを処理するメイン関数"""
+    logging.info(f"Received file: {pdf_file.name if hasattr(pdf_file, 'name') else pdf_file}")
 
     with tempfile.TemporaryDirectory() as temp_dir:
-        # PDFを分割
-        split_pdf_paths = split_pdf(pdf_path, temp_dir)
+        temp_pdf_path = pdf_file.name
+        logging.info(f"Temporary PDF path: {temp_pdf_path}")
+
+        split_pdf_paths = split_pdf(temp_pdf_path, temp_dir)
         logging.info(f"Split PDF paths: {split_pdf_paths}")
+        progress(0.2, desc="PDFを分割中...")
 
-        # 並列処理でOCR変換
         markdown_results = {}
         with ThreadPoolExecutor() as executor:
-            futures = {executor.submit(ocr_pdf_with_gemini, path, model_name): start_page for start_page, path in split_pdf_paths}
+            futures = {executor.submit(ocr_pdf_with_gemini, path, MODEL_NAME): start_page for start_page, path in split_pdf_paths}
+            completed_count = 0
             for future in as_completed(futures):
                 start_page = futures[future]
                 try:
                     result = future.result()
                     markdown_results[start_page] = result
+                    completed_count += 1
+                    progress(0.2 + 0.6 * (completed_count) / len(futures), desc="OCR処理中...")
                 except Exception as e:
                     logging.error(f"Error processing split PDF: {e}")
                     markdown_results[start_page] = f"分割PDFの処理中にエラーが発生しました: {e}"
 
-        # 結果を結合 (開始ページ番号でソート)
+        logging.info(f"Markdown results length: {len(markdown_results)}")
+        progress(0.8, desc="結果を結合中...")
+
         combined_markdown = "\n\n".join(markdown_results[page] for page in sorted(markdown_results.keys()))
+        progress(1.0, desc="完了")
+        time.sleep(0.5)
+
         return combined_markdown
 
 
+def create_interface():
+    with gr.Blocks() as demo:
+        gr.Markdown("# PDF OCR & マークダウン変換ツール")
+        gr.Markdown("PDFをアップロードすると、OCRでテキストを抽出しマークダウン形式に変換します。")
 
-# --- FastAPI 定義 ---
-
-app = FastAPI()
-
-# リクエストモデル
-class PDFRequest(BaseModel):
-    pdf_url: HttpUrl = Field(..., description="URL of the PDF file to be processed.")
-    request_id: Optional[str] = Field(None, description="Optional request ID for tracking.") # 追跡用のID
-
-# レスポンスモデル
-class PDFResponse(BaseModel):
-    request_id: Optional[str] = Field(None, description="Request ID")
-    title: str = Field(..., description="Title extracted from the PDF (if available).")
-    markdown_content: str = Field(..., description="The extracted text in Markdown format.")
-    processing_time: float = Field(..., description="Time taken for processing in seconds.")
-    status: str = Field("success", description="Status of the request (success or failure).")
-    error_message: Optional[str] = Field(None, description="Error message if processing failed.")
-
-async def download_pdf(url: str, temp_dir: str) -> str:
-    """PDFをURLからダウンロードする関数 (非同期)"""
-    async with httpx.AsyncClient() as client:
-        try:
-            response = await client.get(url, follow_redirects=True, timeout=30)  # リダイレクト対応, タイムアウト設定
-            response.raise_for_status()  # エラーチェック
-
-            # 一時ファイルに保存
-            file_extension = os.path.splitext(url)[1] or ".pdf" # 拡張子取得 or デフォルト.pdf
-            with tempfile.NamedTemporaryFile(delete=False, dir=temp_dir, suffix=file_extension) as temp_file:
-                temp_file.write(response.content)
-                temp_file_path = temp_file.name
-            logging.info(f"Downloaded PDF to: {temp_file_path}")
-            return temp_file_path
-
-        except httpx.RequestError as exc:
-            logging.error(f"Request failed: {exc}")
-            raise HTTPException(status_code=400, detail=f"Failed to download PDF from URL: {exc}")
-        except httpx.HTTPStatusError as exc:
-            logging.error(f"HTTP error: {exc}")
-            raise HTTPException(status_code=exc.response.status_code, detail=str(exc))
-        except Exception as e:
-            logging.error(f"Download failed {e}")
-            raise HTTPException(status_code=500, detail=str(e))
-
-
-@app.post("/process-pdf/", response_model=PDFResponse)
-async def process_pdf_endpoint(request: PDFRequest, background_tasks: BackgroundTasks):
-    """PDFを処理するエンドポイント"""
-    start_time = time.time()
-    request_id = request.request_id or str(time.time()) # リクエストIDがない場合は生成
+        with gr.Row():
+            pdf_input = gr.File(label="PDFファイルをアップロード", file_types=[".pdf"])
 
-    with tempfile.TemporaryDirectory() as temp_dir:
-        try:
-            # PDFをダウンロード (非同期)
-            pdf_path = await download_pdf(str(request.pdf_url), temp_dir)
+        with gr.Row():
+            convert_btn = gr.Button("変換開始", variant="primary",
+                                    elem_id="convert-button")
 
-            # PDF処理 (同期処理をバックグラウンドで実行)
-            markdown_content = await background_tasks.add_task(process_pdf, pdf_path, MODEL_NAME)
+        with gr.Row():
+            markdown_output = gr.Textbox(label="変換結果", lines=10,
+                                         max_lines=20)
 
-            # PDFからタイトルを取得 (可能であれば)
-            try:
-                with fitz.open(pdf_path) as doc:
-                    title = doc.metadata.get("title", "Untitled")
-            except Exception:
-                title = "Untitled"
+        with gr.Row():
+            copy_btn = gr.Button("クリップボードにコピー")
+            download_btn = gr.Button("ダウンロード")
 
+        js_code = gr.HTML(
+            """
+            <script>
+            function styleButton() {
+              document.getElementById('convert-button').style.backgroundColor = 'orange';
+            }
+            </script>
+            """,
+            visible=False,
+        )
 
-            end_time = time.time()
-            processing_time = end_time - start_time
+        pdf_input.upload(None, [], [], js="styleButton")
+
+        convert_btn.click(
+            fn=process_pdf,
+            inputs=pdf_input,
+            outputs=markdown_output
+        )
+
+        copy_btn.click(
+            None,
+            markdown_output,
+            [],
+            js=f"(x) => {{ navigator.clipboard.writeText(x); }}",
+        )
+
+        download_btn.click(
+            None,
+            markdown_output,
+            [],
+            js=f"""(x) =>{{
+            const blob = new Blob([x], {{type: 'text/markdown;charset=utf-8'}});
+            const url = URL.createObjectURL(blob);
+            const a = document.createElement('a');
+            a.href = url;
+            a.download = 'converted.md';
+            document.body.appendChild(a);
+            a.click();
+            document.body.removeChild(a);
+            URL.revokeObjectURL(url);
+          }}"""
+        )
+    return demo
 
-            return PDFResponse(
-                request_id=request_id,
-                title=title,
-                markdown_content=markdown_content,
-                processing_time=processing_time
-            )
 
-        except HTTPException as e: # HTTPレベルエラー
-            logging.exception("HTTPException during processing:")
-            end_time = time.time()
-            return JSONResponse(
-                status_code=e.status_code,
-                content=PDFResponse(
-                    request_id = request_id,
-                    title="Error",
-                    markdown_content="",
-                    processing_time=end_time - start_time,
-                    status="failure",
-                    error_message=e.detail
-                    ).dict()
-                )
-        except Exception as e: # その他のエラー
-            logging.exception("Exception during processing:")
-            end_time = time.time()
-            return JSONResponse(
-                status_code=500,
-                content=PDFResponse(
-                    request_id=request_id,
-                    title="Error",
-                    markdown_content="",
-                    processing_time=end_time - start_time,
-                    status="failure",
-                    error_message=str(e)
-                ).dict()
-            )
+if __name__ == "__main__":
+    demo = create_interface()
+    demo.launch()

requirements.txt

@@ -1,6 +1,3 @@
-fastapi
-uvicorn
-httpx
+gradio==4.44.1
 PyMuPDF
-google-generativeai
-pydantic
+google-generativeai