fix: process pdf once

app-ocr.py

@@ -1,146 +1,82 @@
 import base64
-import io
 import os
-import shutil
-import time
-import uuid
-from pathlib import Path
 
 import gradio as gr
-from modelscope import AutoModel, AutoTokenizer
+import spaces
+import torch
+from transformers import AutoModel, AutoTokenizer
 
-UPLOAD_FOLDER = "./uploads"
-RESULTS_FOLDER = "./results"
+model_name = "ucaslcl/GOT-OCR2_0"
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+model = AutoModel.from_pretrained(model_name, trust_remote_code=True, device_map=device)
+model = model.eval().to(device)
 
-tokenizer = AutoTokenizer.from_pretrained("stepfun-ai/GOT-OCR2_0", trust_remote_code=True)
-model = AutoModel.from_pretrained("stepfun-ai/GOT-OCR2_0", trust_remote_code=True, low_cpu_mem_usage=True, device_map="cuda", use_safetensors=True)
-model = model.eval().cuda()
 
-for folder in [UPLOAD_FOLDER, RESULTS_FOLDER]:
-    if not os.path.exists(folder):
-        os.makedirs(folder)
+@spaces.GPU()
+def ocr_process(image, got_mode, ocr_color="", ocr_box="", progress=gr.Progress()):
+    if image is None:
+        return "错误:未提供图片"
 
+    try:
+        image_path = image
+        result_path = f"{os.path.splitext(image_path)[0]}_result.html"
+
+        progress(0, desc="开始处理...")
+
+        if "plain" in got_mode:
+            progress(0.3, desc="执行OCR识别...")
+            if "multi-crop" in got_mode:
+                res = model.chat_crop(tokenizer, image_path, ocr_type="ocr")
+            else:
+                res = model.chat(tokenizer, image_path, ocr_type="ocr", ocr_box=ocr_box, ocr_color=ocr_color)
+            progress(1, desc="处理完成")
+            return res
+        elif "format" in got_mode:
+            progress(0.3, desc="执行OCR识别...")
+            if "multi-crop" in got_mode:
+                res = model.chat_crop(tokenizer, image_path, ocr_type="format", render=True, save_render_file=result_path)
+            else:
+                res = model.chat(tokenizer, image_path, ocr_type="format", ocr_box=ocr_box, ocr_color=ocr_color, render=True, save_render_file=result_path)
+
+            progress(0.7, desc="生成结果...")
+            if os.path.exists(result_path):
+                with open(result_path, "r", encoding="utf-8") as f:
+                    html_content = f.read()
+                encoded_html = base64.b64encode(html_content.encode("utf-8")).decode("utf-8")
+                data_uri = f"data:text/html;charset=utf-8;base64,{encoded_html}"
+                preview = f'<iframe src="{data_uri}" width="100%" height="600px"></iframe>'
+                download_link = f'<a href="{data_uri}" download="result.html">下载完整结果</a>'
+                progress(1, desc="处理完成")
+                return f"{download_link}\n\n{preview}"
+
+        return "错误: 未知的OCR模式"
+    except Exception as e:
+        return f"错误: {str(e)}"
 
-def image_to_base64(image):
-    buffered = io.BytesIO()
-    image.save(buffered, format="PNG")
-    return base64.b64encode(buffered.getvalue()).decode()
 
+with gr.Blocks() as demo:
+    gr.Markdown("# OCR 图像识别")
 
-def run_GOT(image, got_mode, fine_grained_mode="", ocr_color="", ocr_box=""):
-    unique_id = str(uuid.uuid4())
-    image_path = os.path.join(UPLOAD_FOLDER, f"{unique_id}.png")
-    result_path = os.path.join(RESULTS_FOLDER, f"{unique_id}.html")
+    with gr.Row():
+        image_input = gr.Image(type="filepath", label="上传图片")
 
-    shutil.copy(image, image_path)
+    got_mode = gr.Dropdown(
+        choices=["plain texts OCR", "format texts OCR", "plain multi-crop OCR", "format multi-crop OCR", "plain fine-grained OCR", "format fine-grained OCR"],
+        label="OCR模式",
+        value="plain texts OCR",
+    )
 
-    try:
-        if got_mode == "plain texts OCR":
-            res = model.chat(tokenizer, image_path, ocr_type="ocr")
-            return res, None
-        elif got_mode == "format texts OCR":
-            res = model.chat(tokenizer, image_path, ocr_type="format", render=True, save_render_file=result_path)
-        elif got_mode == "plain multi-crop OCR":
-            res = model.chat_crop(tokenizer, image_path, ocr_type="ocr")
-            return res, None
-        elif got_mode == "format multi-crop OCR":
-            res = model.chat_crop(tokenizer, image_path, ocr_type="format", render=True, save_render_file=result_path)
-        elif got_mode == "plain fine-grained OCR":
-            res = model.chat(tokenizer, image_path, ocr_type="ocr", ocr_box=ocr_box, ocr_color=ocr_color)
-            return res, None
-        elif got_mode == "format fine-grained OCR":
-            res = model.chat(tokenizer, image_path, ocr_type="format", ocr_box=ocr_box, ocr_color=ocr_color, render=True, save_render_file=result_path)
-
-        # res_markdown = f"$$ {res} $$"
-        res_markdown = res
-
-        if "format" in got_mode and os.path.exists(result_path):
-            with open(result_path, "r") as f:
-                html_content = f.read()
-            encoded_html = base64.b64encode(html_content.encode("utf-8")).decode("utf-8")
-            iframe_src = f"data:text/html;base64,{encoded_html}"
-            iframe = f'<iframe src="{iframe_src}" width="100%" height="600px"></iframe>'
-            download_link = f'<a href="data:text/html;base64,{encoded_html}" download="result_{unique_id}.html">Download Full Result</a>'
-            return res_markdown, f"{download_link}<br>{iframe}"
-        else:
-            return res_markdown, None
-    except Exception as e:
-        return f"Error: {str(e)}", None
-    finally:
-        if os.path.exists(image_path):
-            os.remove(image_path)
-
-
-def task_update(task):
-    if "fine-grained" in task:
-        return [
-            gr.update(visible=True),
-            gr.update(visible=False),
-            gr.update(visible=False),
-        ]
-    else:
-        return [
-            gr.update(visible=False),
-            gr.update(visible=False),
-            gr.update(visible=False),
-        ]
-
-
-def fine_grained_update(task):
-    if task == "box":
-        return [
-            gr.update(visible=False, value=""),
-            gr.update(visible=True),
-        ]
-    elif task == "color":
-        return [
-            gr.update(visible=True),
-            gr.update(visible=False, value=""),
-        ]
-
-
-def cleanup_old_files():
-    current_time = time.time()
-    for folder in [UPLOAD_FOLDER, RESULTS_FOLDER]:
-        for file_path in Path(folder).glob("*"):
-            if current_time - file_path.stat().st_mtime > 3600:  # 1 hour
-                file_path.unlink()
+    with gr.Row():
+        ocr_color = gr.Textbox(label="OCR颜色 (仅用于fine-grained模式)")
+        ocr_box = gr.Textbox(label="OCR边界框 (仅用于fine-grained模式)")
 
+    submit_button = gr.Button("开始OCR识别")
 
-with gr.Blocks() as demo:
-    with gr.Row():
-        with gr.Column():
-            image_input = gr.Image(type="filepath", label="上传图片")
-            task_dropdown = gr.Dropdown(
-                choices=[
-                    "plain texts OCR",
-                    "format texts OCR",
-                    "plain multi-crop OCR",
-                    "format multi-crop OCR",
-                    "plain fine-grained OCR",
-                    "format fine-grained OCR",
-                ],
-                label="选择GOT模式",
-                value="plain texts OCR",
-            )
-            fine_grained_dropdown = gr.Dropdown(choices=["box", "color"], label="fine-grained type", visible=False)
-            color_dropdown = gr.Dropdown(choices=["red", "green", "blue"], label="color list", visible=False)
-            box_input = gr.Textbox(label="input box: [x1,y1,x2,y2]", placeholder="e.g., [0,0,100,100]", visible=False)
-            submit_button = gr.Button("Submit")
-
-        with gr.Column():
-            ocr_result = gr.Textbox(label="GOT output")
-
-    with gr.Column():
-        gr.Markdown("**如果选择带格式的模式，mathpix结果将自动呈现如下:**")
-        html_result = gr.HTML(label="rendered html", show_label=True)
-
-    task_dropdown.change(task_update, inputs=[task_dropdown], outputs=[fine_grained_dropdown, color_dropdown, box_input])
-    fine_grained_dropdown.change(fine_grained_update, inputs=[fine_grained_dropdown], outputs=[color_dropdown, box_input])
-
-    submit_button.click(run_GOT, inputs=[image_input, task_dropdown, fine_grained_dropdown, color_dropdown, box_input], outputs=[ocr_result, html_result])
+    output = gr.HTML(label="识别结果")
+
+    submit_button.click(ocr_process, inputs=[image_input, got_mode, ocr_color, ocr_box], outputs=output)
 
 if __name__ == "__main__":
-    cleanup_old_files()
     demo.launch()

app.py

@@ -1,10 +1,12 @@
 import base64
 import os
-import time
+import tempfile
 
+import fitz
 import gradio as gr
 import spaces
 import torch
+from PIL import Image, ImageEnhance
 from transformers import AutoModel, AutoTokenizer
 
 model_name = "ucaslcl/GOT-OCR2_0"
@@ -15,53 +17,95 @@ model = AutoModel.from_pretrained(model_name, trust_remote_code=True, device_map
 model = model.eval().to(device)
 
 
+def pdf_to_images(pdf_path):
+    images = []
+    pdf_document = fitz.open(pdf_path)
+    for page_num in range(len(pdf_document)):
+        page = pdf_document.load_page(page_num)
+        zoom = 10  # 增加缩放比例到10
+        mat = fitz.Matrix(zoom, zoom)
+        pix = page.get_pixmap(matrix=mat, alpha=False)
+        img = Image.frombytes("RGB", [pix.width, pix.height], pix.samples)
+
+        # 增对比度
+        enhancer = ImageEnhance.Contrast(img)
+        img = enhancer.enhance(1.5)  # 增加50%的对比度
+
+        images.append(img)
+    pdf_document.close()
+    return images
+
+
 @spaces.GPU()
-def ocr_process(image, got_mode, ocr_color="", ocr_box="", progress=gr.Progress()):
-    if image is None:
-        return "错误:未提供图片"
+def ocr_process(file, got_mode, ocr_color="", ocr_box="", progress=gr.Progress()):
+    if file is None:
+        return "错误:未提供文件"
 
     try:
-        image_path = image
-        result_path = f"{os.path.splitext(image_path)[0]}_result.html"
-
         progress(0, desc="开始处理...")
 
-        if "plain" in got_mode:
-            progress(0.3, desc="执行OCR识别...")
-            if "multi-crop" in got_mode:
-                res = model.chat_crop(tokenizer, image_path, ocr_type="ocr")
-            else:
-                res = model.chat(tokenizer, image_path, ocr_type="ocr", ocr_box=ocr_box, ocr_color=ocr_color)
-            progress(1, desc="处理完成")
-            return res
-        elif "format" in got_mode:
-            progress(0.3, desc="执行OCR识别...")
-            if "multi-crop" in got_mode:
-                res = model.chat_crop(tokenizer, image_path, ocr_type="format", render=True, save_render_file=result_path)
+        with tempfile.TemporaryDirectory() as temp_dir:
+            if file.name.lower().endswith(".pdf"):
+                pdf_path = os.path.join(temp_dir, "temp.pdf")
+                with open(pdf_path, "wb") as f:
+                    f.write(file.read())
+
+                images = pdf_to_images(pdf_path)
+                num_pages = len(images)
+                results = []
+
+                for i, image in enumerate(images):
+                    progress((i + 1) / num_pages, desc=f"处理第 {i+1}/{num_pages} 页...")
+                    img_path = os.path.join(temp_dir, f"page_{i+1}.png")
+                    image.save(img_path, "PNG")
+
+                    result = process_single_image(img_path, got_mode, ocr_color, ocr_box)
+                    results.append(f"第 {i+1} 页结果:\n{result}")
+
+                final_result = "\n\n".join(results)
             else:
-                res = model.chat(tokenizer, image_path, ocr_type="format", ocr_box=ocr_box, ocr_color=ocr_color, render=True, save_render_file=result_path)
-
-            progress(0.7, desc="生成结果...")
-            if os.path.exists(result_path):
-                with open(result_path, "r", encoding="utf-8") as f:
-                    html_content = f.read()
-                encoded_html = base64.b64encode(html_content.encode("utf-8")).decode("utf-8")
-                data_uri = f"data:text/html;charset=utf-8;base64,{encoded_html}"
-                preview = f'<iframe src="{data_uri}" width="100%" height="600px"></iframe>'
-                download_link = f'<a href="{data_uri}" download="result.html">下载完整结果</a>'
-                progress(1, desc="处理完成")
-                return f"{download_link}\n\n{preview}"
-
-        return "错误: 未知的OCR模式"
+                img_path = os.path.join(temp_dir, "temp_image.png")
+                with open(img_path, "wb") as f:
+                    f.write(file.read())
+                final_result = process_single_image(img_path, got_mode, ocr_color, ocr_box)
+
+        progress(1, desc="处理完成")
+        return final_result
     except Exception as e:
         return f"错误: {str(e)}"
 
 
+def process_single_image(image_path, got_mode, ocr_color, ocr_box):
+    result_path = f"{os.path.splitext(image_path)[0]}_result.html"
+
+    if "plain" in got_mode:
+        if "multi-crop" in got_mode:
+            res = model.chat_crop(tokenizer, image_path, ocr_type="ocr")
+        else:
+            res = model.chat(tokenizer, image_path, ocr_type="ocr", ocr_box=ocr_box, ocr_color=ocr_color)
+        return res
+    elif "format" in got_mode:
+        if "multi-crop" in got_mode:
+            res = model.chat_crop(tokenizer, image_path, ocr_type="format", render=True, save_render_file=result_path)
+        else:
+            res = model.chat(tokenizer, image_path, ocr_type="format", ocr_box=ocr_box, ocr_color=ocr_color, render=True, save_render_file=result_path)
+
+        if os.path.exists(result_path):
+            with open(result_path, "r", encoding="utf-8") as f:
+                html_content = f.read()
+            encoded_html = base64.b64encode(html_content.encode("utf-8")).decode("utf-8")
+            data_uri = f"data:text/html;charset=utf-8;base64,{encoded_html}"
+            preview = f'<iframe src="{data_uri}" width="100%" height="600px"></iframe>'
+            download_link = f'<a href="{data_uri}" download="result.html">下载完整结果</a>'
+            return f"{download_link}\n\n{preview}\n\n识别结果：\n{res}"
+
+    return "错误: 未知的OCR模式"
+
+
 with gr.Blocks() as demo:
     gr.Markdown("# OCR 图像识别")
 
-    with gr.Row():
-        image_input = gr.Image(type="filepath", label="上传图片")
+    file_input = gr.File(label="上传PDF或图片文件")
 
     got_mode = gr.Dropdown(
         choices=["plain texts OCR", "format texts OCR", "plain multi-crop OCR", "format multi-crop OCR", "plain fine-grained OCR", "format fine-grained OCR"],
@@ -77,7 +121,7 @@ with gr.Blocks() as demo:
 
     output = gr.HTML(label="识别结果")
 
-    submit_button.click(ocr_process, inputs=[image_input, got_mode, ocr_color, ocr_box], outputs=output)
+    submit_button.click(ocr_process, inputs=[file_input, got_mode, ocr_color, ocr_box], outputs=output)
 
 if __name__ == "__main__":
     demo.launch()