fix: process pdf once

app-pdf.py

@@ -1,32 +1,20 @@
+import base64
 import os
-import shutil
-import uuid
+import tempfile
 
-import fitz  # PyMuPDF
+import fitz
 import gradio as gr
+import spaces
 import torch
 from PIL import Image, ImageEnhance
-from transformers import AutoConfig, AutoModel, AutoTokenizer
+from transformers import AutoModel, AutoTokenizer
 
-from got_ocr import got_ocr
-
-# 初始化模型和分词器
-# model_name = "stepfun-ai/GOT-OCR2_0"
 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)
-config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
-model = AutoModel.from_pretrained(model_name, trust_remote_code=True, low_cpu_mem_usage=True, device_map="cuda", use_safetensors=True)
+model = AutoModel.from_pretrained(model_name, trust_remote_code=True, device_map=device)
 model = model.eval().to(device)
-model.config.pad_token_id = tokenizer.eos_token_id
-
-UPLOAD_FOLDER = "./uploads"
-RESULTS_FOLDER = "./results"
-
-# 确保必要的文件夹存在
-os.makedirs(UPLOAD_FOLDER, exist_ok=True)
-os.makedirs(RESULTS_FOLDER, exist_ok=True)
 
 
 def pdf_to_images(pdf_path):
@@ -34,8 +22,7 @@ def pdf_to_images(pdf_path):
     pdf_document = fitz.open(pdf_path)
     for page_num in range(len(pdf_document)):
         page = pdf_document.load_page(page_num)
-        # 进一步增加分辨率和缩放比例
-        zoom = 10  # 增加缩放比例到4
+        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)
@@ -49,116 +36,87 @@ def pdf_to_images(pdf_path):
     return images
 
 
-def process_pdf(pdf_file):
-    if pdf_file is None:
-        return None
-
-    temp_pdf_path = os.path.join(UPLOAD_FOLDER, f"{uuid.uuid4()}.pdf")
-
-    # 使用 shutil 复制上传的件到临时位置
-    shutil.copy(pdf_file.name, temp_pdf_path)
-
-    images = pdf_to_images(temp_pdf_path)
-    os.remove(temp_pdf_path)
+@spaces.GPU()
+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"
 
-    # 将图像保存为临时文件并返回文件路径列表
-    image_paths = []
-    for i, img in enumerate(images):
-        img_path = os.path.join(RESULTS_FOLDER, f"page_{i+1}.png")
-        img.save(img_path, "PNG")
-        image_paths.append(img_path)
+    try:
+        progress(0, desc="开始处理...")
 
-    return image_paths
+        with tempfile.TemporaryDirectory() as temp_dir:
+            file_path = file.name  # 使用文件的原始路径
 
+            if file_path.lower().endswith(".pdf"):
+                images = pdf_to_images(file_path)
+                num_pages = len(images)
+                results = []
 
-def on_image_select(evt: gr.SelectData):
-    if evt.index is not None:
-        return evt.index
-    return None
+                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}")
 
-# 更新perform_ocr函数的输入参数
-def perform_ocr(selected_index, image_gallery, got_mode, fine_grained_type, color, box):
-    if selected_index is None or len(image_gallery) == 0:
-        return "请先选择一张图片"
+                final_result = "\n\n".join(results)
+            else:
+                final_result = process_single_image(file_path, got_mode, ocr_color, ocr_box)
 
-    selected_image = image_gallery[selected_index][0]
+        progress(1, desc="处理完成")
+        return final_result
+    except Exception as e:
+        return f"错误: {str(e)}"
 
-    # 根据选择的任务和参数调用GOT OCR
-    ocr_color = color if fine_grained_type == "color" else ""
-    ocr_box = box if fine_grained_type == "box" else ""
-
-    result, html_content = got_ocr(
-        model,
-        tokenizer,
-        selected_image,
-        got_mode=got_mode,
-        fine_grained_mode=fine_grained_type,
-        ocr_color=ocr_color,
-        ocr_box=ocr_box,
-    )
-
-    if html_content:
-        iframe_src = f"data:text/html;base64,{html_content}"
-        iframe = f'<iframe src="{iframe_src}" width="100%" height="600px"></iframe>'
-        download_link = f'<a href="data:text/html;base64,{html_content}" download="result.html">下载完整结果</a>'
-        return gr.Markdown(f"{download_link}\n\n{iframe}")
-    else:
-        return gr.Markdown(result)
 
+def process_single_image(image_path, got_mode, ocr_color, ocr_box):
+    result_path = f"{os.path.splitext(image_path)[0]}_result.html"
 
-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)]
+    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}"
 
-def fine_grained_update(fine_grained_type):
-    if fine_grained_type == "color":
-        return [gr.update(visible=True), gr.update(visible=False)]
-    elif fine_grained_type == "box":
-        return [gr.update(visible=False), gr.update(visible=True)]
-    else:
-        return [gr.update(visible=False), gr.update(visible=False)]
+    return "错误: 未知的OCR模式"
 
 
 with gr.Blocks() as demo:
-    pdf_input = gr.File(label="上传PDF文件")
-    image_gallery = gr.Gallery(
-        label="PDF页面预览",
-        columns=3,
-        height=600,
-        object_fit="contain",
-        preview=True,
-    )
-    selected_index = gr.State(None)
-    pdf_input.upload(fn=process_pdf, inputs=pdf_input, outputs=image_gallery)
-    image_gallery.select(fn=on_image_select, inputs=[], outputs=selected_index)
+    gr.Markdown("# OCR 图像识别")
+
+    file_input = gr.File(label="上传PDF或图片文件")
 
-    ####################
-    task_dropdown = gr.Dropdown(
+    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="选择GOT模式",
-        value="format texts OCR",
+        label="OCR模式",
+        value="plain texts OCR",
     )
-    fine_grained_dropdown = gr.Dropdown(choices=["box", "color"], label="fine-grained类型", visible=False)
-    color_dropdown = gr.Dropdown(choices=["red", "green", "blue"], label="颜色列表", visible=False)
-    box_input = gr.Textbox(label="输入框: [x1,y1,x2,y2]", placeholder="例如: [0,0,100,100]", visible=False)
 
-    ocr_button = gr.Button("开始OCR识别")
-    ocr_result = gr.HTML(label="OCR结果")  # 将Textbox更改为HTML组件
+    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)")
 
-    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 = gr.Button("开始OCR识别")
 
-    # 更新ocr_button的click事件，传递所有必要的参数
-    ocr_button.click(
-        fn=perform_ocr,
-        inputs=[selected_index, image_gallery, task_dropdown, fine_grained_dropdown, color_dropdown, box_input],
-        outputs=ocr_result,
-    )
+    output = gr.HTML(label="识别结果")
 
+    submit_button.click(ocr_process, inputs=[file_input, got_mode, ocr_color, ocr_box], outputs=output)
 
 if __name__ == "__main__":
     demo.launch()

app.py

@@ -37,35 +37,38 @@ def pdf_to_images(pdf_path):
 
 
 @spaces.GPU()
-def ocr_process(file, got_mode, ocr_color="", ocr_box="", progress=gr.Progress()):
+def convert_pdf_to_images(file):
     if file is None:
-        return "错误:未提供文件"
+        return "错误:未提供文件", None
 
     try:
-        progress(0, desc="开始处理...")
+        if not file.name.lower().endswith(".pdf"):
+            return "错误:请上传PDF文件", None
 
-        with tempfile.TemporaryDirectory() as temp_dir:
-            file_path = file.name  # 使用文件的原始路径
+        images = pdf_to_images(file.name)
+        image_paths = []
 
-            if file_path.lower().endswith(".pdf"):
-                images = pdf_to_images(file_path)
-                num_pages = len(images)
-                results = []
+        with tempfile.TemporaryDirectory() as temp_dir:
+            for i, image in enumerate(images):
+                img_path = os.path.join(temp_dir, f"page_{i+1}.png")
+                image.save(img_path, "PNG")
+                image_paths.append(img_path)
 
-                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")
+        return "PDF转换为图片成功", image_paths
+    except Exception as e:
+        return f"错误: {str(e)}", None
 
-                    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:
-                final_result = process_single_image(file_path, got_mode, ocr_color, ocr_box)
+@spaces.GPU()
+def ocr_process(image, got_mode, ocr_color="", ocr_box="", progress=gr.Progress()):
+    if image is None:
+        return "错误:未选择图片"
 
+    try:
+        progress(0, desc="开始处理...")
+        result = process_single_image(image, got_mode, ocr_color, ocr_box)
         progress(1, desc="处理完成")
-        return final_result
+        return result
     except Exception as e:
         return f"错误: {str(e)}"
 
@@ -92,7 +95,7 @@ def process_single_image(image_path, got_mode, ocr_color, ocr_box):
             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 f"{download_link}\n\n{preview}"
 
     return "错误: 未知的OCR模式"
 
@@ -100,23 +103,36 @@ def process_single_image(image_path, got_mode, ocr_color, ocr_box):
 with gr.Blocks() as demo:
     gr.Markdown("# OCR 图像识别")
 
-    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"],
-        label="OCR模式",
-        value="plain texts OCR",
-    )
-
-    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")
-
-    output = gr.HTML(label="识别结果")
-
-    submit_button.click(ocr_process, inputs=[file_input, got_mode, ocr_color, ocr_box], outputs=output)
+    with gr.Tab("PDF转图片"):
+        pdf_input = gr.File(label="上传PDF文件")
+        convert_button = gr.Button("转换PDF为图片")
+        pdf_output = gr.Textbox(label="转换结果")
+        image_gallery = gr.Gallery(label="图片预览").style(grid=3)
+
+    with gr.Tab("OCR处理"):
+        image_input = gr.Image(label="选择或上传图片")
+        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",
+        )
+
+        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)")
+
+        ocr_button = gr.Button("开始OCR识别")
+        ocr_output = gr.HTML(label="识别结果")
+
+    convert_button.click(convert_pdf_to_images, inputs=[pdf_input], outputs=[pdf_output, image_gallery])
+    ocr_button.click(ocr_process, inputs=[image_input, got_mode, ocr_color, ocr_box], outputs=ocr_output)
 
 if __name__ == "__main__":
     demo.launch()