fix: process pdf once

.gitignore

@@ -1,2 +1,3 @@
 results/*
 uploads/*
+__pycache__

app-pdf.py

@@ -0,0 +1,164 @@
+import os
+import shutil
+import uuid
+
+import fitz  # PyMuPDF
+import gradio as gr
+import torch
+from PIL import Image, ImageEnhance
+from transformers import AutoConfig, 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 = 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):
+    images = []
+    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
+        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
+
+
+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)
+
+    # 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)
+
+    return image_paths
+
+
+def on_image_select(evt: gr.SelectData):
+    if evt.index is not None:
+        return evt.index
+    return None
+
+
+# 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"
+
+    selected_image = image_gallery[selected_index][0]
+
+    # 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 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(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)]
+
+
+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)
+
+    ####################
+    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="format 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
+
+    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])
+
+    # 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,
+    )
+
+
+if __name__ == "__main__":
+    demo.launch()

app.py

@@ -1,3 +1,4 @@
+import base64
 import os
 import shutil
 import uuid
@@ -78,33 +79,37 @@ def on_image_select(evt: gr.SelectData):
 
 
 # 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"
-
-    selected_image = image_gallery[selected_index][0]
-
-    # 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,
-    )
+def perform_ocr(image_gallery, got_mode, fine_grained_type, color, box):
+    if len(image_gallery) == 0:
+        return "请先上传PDF文件", None
 
-    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)
+    results = []
+    progress = gr.Progress()
+
+    for i, image_info in enumerate(progress.tqdm(image_gallery)):
+        selected_image = image_info[0]
+
+        ocr_color = color if fine_grained_type == "color" else ""
+        ocr_box = box if fine_grained_type == "box" else ""
+
+        result, _ = got_ocr(
+            model,
+            tokenizer,
+            selected_image,
+            got_mode=got_mode,
+            fine_grained_mode=fine_grained_type,
+            ocr_color=ocr_color,
+            ocr_box=ocr_box,
+        )
+        results.append(f"第 {i+1} 页结果:\n{result}\n\n")
+
+    combined_result = "".join(results)
+
+    # 生成下载链接
+    encoded_result = base64.b64encode(combined_result.encode("utf-8")).decode("utf-8")
+    download_link = f'<a href="data:text/plain;base64,{encoded_result}" download="ocr_result.txt">下载完整OCR结果</a>'
+
+    return gr.Markdown(f"{download_link}\n\n{combined_result[:1000]}..."), combined_result
 
 
 def task_update(task):
@@ -132,11 +137,8 @@ with gr.Blocks() as demo:
         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)
 
-    ####################
     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",
@@ -147,18 +149,17 @@ with gr.Blocks() as demo:
     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
+    ocr_result = gr.Markdown(label="OCR结果预览")
+    full_result = gr.State()
 
     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])
 
-    # 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,
+        inputs=[image_gallery, task_dropdown, fine_grained_dropdown, color_dropdown, box_input],
+        outputs=[ocr_result, full_result],
     )
 
-
 if __name__ == "__main__":
     demo.launch()