add HTML renders (#2)

- add html rendering (eaedd343fb7a8659e5b2bc25015121ce6f839c01)
- add latex display (fc2b70dc3a6292b1061737e74496aa30814fae38)


Co-authored-by: Joseph Pollack <[email protected]>

app.py

@@ -4,36 +4,72 @@ from transformers import AutoModel, AutoTokenizer
 from PIL import Image
 import numpy as np
 import os
+import base64
+import io
+import uuid
+import tempfile
+import time
+import shutil
+from pathlib import Path
 
 tokenizer = AutoTokenizer.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True)
 model = AutoModel.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True, low_cpu_mem_usage=True, device_map='cuda', use_safetensors=True)
 model = model.eval().cuda()
-html_file = './demo.html'
+
+UPLOAD_FOLDER = "./uploads"
+RESULTS_FOLDER = "./results"
+
+for folder in [UPLOAD_FOLDER, RESULTS_FOLDER]:
+    if not os.path.exists(folder):
+        os.makedirs(folder)
+
+def image_to_base64(image):
+    buffered = io.BytesIO()
+    image.save(buffered, format="PNG")
+    return base64.b64encode(buffered.getvalue()).decode()
 
 @spaces.GPU
-def run_GOT(image_array, got_mode, fine_grained_mode="", ocr_color="", ocr_box=""):
-    image = image_array
-    if got_mode == "plain texts OCR":
-        res = model.chat(tokenizer, image, ocr_type='ocr')
-    elif got_mode == "format texts OCR":
-        res = model.chat(tokenizer, image, ocr_type='format', render=True, save_render_file=html_file)
-    elif got_mode == "plain multi-crop OCR":
-        res = model.chat_crop(tokenizer, image, ocr_type='ocr')
-    elif got_mode == "format multi-crop OCR":
-        res = model.chat_crop(tokenizer, image, ocr_type='format', render=True, save_render_file=html_file)
-    elif got_mode == "plain fine-grained OCR":
-        res = model.chat(tokenizer, image, ocr_type='ocr', ocr_box=ocr_box, ocr_color=ocr_color)
-    elif got_mode == "format fine-grained OCR":
-        res = model.chat(tokenizer, image, ocr_type='format', ocr_box=ocr_box, ocr_color=ocr_color, render=True, save_render_file=html_file)
+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")
+    
+    shutil.copy(image, image_path)
+    
+    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"$$ {res} $$"
 
-    # print("res:\n", res)
-    if "format" in got_mode:
-        with open(html_file, 'r') as f:
-            demo_html = f.read()
-        # print("demo_html: \n", demo_html)
-        print(os.path.abspath(html_file))
-        return res, demo_html
-    return res, None
+        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:
@@ -61,6 +97,12 @@ def fine_grained_update(task):
             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()
 
 title_html = """
 <h2> <span class="gradient-text" id="text">General OCR Theory</span><span class="plain-text">: Towards OCR-2.0 via a Unified End-to-end Model</span></h2>
@@ -75,7 +117,6 @@ with gr.Blocks() as demo:
     "🔥🔥🔥This is the official online demo of GOT-OCR-2.0 model!!!"
     
     ### Demo Guidelines
-
     You need to upload your image below and choose one mode of GOT, then click "Submit" to run GOT model. More characters will result in longer wait times.
     - **plain texts OCR & format texts OCR**: The two modes are for the image-level OCR.
     - **plain multi-crop OCR & format multi-crop OCR**: For images with more complex content, you can achieve higher-quality results with these modes.
@@ -115,12 +156,10 @@ with gr.Blocks() as demo:
             submit_button = gr.Button("Submit")
         
         with gr.Column():
-            ocr_result = gr.Textbox(label="GOT output")
-
+            ocr_result = gr.Markdown(label="GOT output")
 
     with gr.Column():
-        html_result = gr.HTML(
-                label="rendered html", show_label=True)
+        html_result = gr.HTML(label="rendered html", show_label=True)
     
     gr.Examples(
         examples=[
@@ -134,7 +173,7 @@ with gr.Blocks() as demo:
         ],
         inputs=[image_input, task_dropdown, fine_grained_dropdown, color_dropdown, box_input],
         outputs=[ocr_result, html_result],
-        fn = run_GOT,
+        fn=run_GOT,
         label="examples",
     )
 
@@ -155,4 +194,6 @@ with gr.Blocks() as demo:
         outputs=[ocr_result, html_result]
     )
 
-demo.launch()
+if __name__ == "__main__":
+    cleanup_old_files()
+    demo.launch()

requirements.txt

@@ -6,4 +6,6 @@ tiktoken
 verovio
 opencv-python
 accelerate
-numpy==1.26.4
+numpy==1.26.4
+shutils
+pillow