update app

app.py

@@ -11,9 +11,7 @@ from typing import Iterable
 import gradio as gr
 import spaces
 import torch
-import numpy as np
 from PIL import Image
-import cv2
 
 from transformers import (
     Qwen2_5_VLForConditionalGeneration,
@@ -142,27 +140,6 @@ model_o = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
-def downsample_video(video_path):
-    """
-    Downsamples the video to evenly spaced frames.
-    Each frame is returned as a PIL image along with its timestamp.
-    """
-    vidcap = cv2.VideoCapture(video_path)
-    total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
-    fps = vidcap.get(cv2.CAP_PROP_FPS)
-    frames = []
-    frame_indices = np.linspace(0, total_frames - 1, min(total_frames, 10), dtype=int)
-    for i in frame_indices:
-        vidcap.set(cv2.CAP_PROP_POS_FRAMES, i)
-        success, image = vidcap.read()
-        if success:
-            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
-            pil_image = Image.fromarray(image)
-            timestamp = round(i / fps, 2)
-            frames.append((pil_image, timestamp))
-    vidcap.release()
-    return frames
-
 @spaces.GPU
 def generate_image(model_name: str, text: str, image: Image.Image,
                    max_new_tokens: int = 1024,
@@ -217,72 +194,7 @@ def generate_image(model_name: str, text: str, image: Image.Image,
         time.sleep(0.01)
         yield buffer, buffer
 
-@spaces.GPU
-def generate_video(model_name: str, text: str, video_path: str,
-                   max_new_tokens: int = 1024,
-                   temperature: float = 0.6,
-                   top_p: float = 0.9,
-                   top_k: int = 50,
-                   repetition_penalty: float = 1.2):
-    """
-    Generates responses using the selected model for video input.
-    Yields raw text and Markdown-formatted text.
-    """
-    if model_name == "docscopeOCR-7B-050425-exp":
-        processor, model = processor_m, model_m
-    elif model_name == "coreOCR-7B-050325-preview":
-        processor, model = processor_x, model_x
-    elif model_name == "MonkeyOCR-Recognition":
-        processor, model = processor_g, model_g
-    elif model_name == "Camel-Doc-OCR-080125(v2)":
-        processor, model = processor_o, model_o
-    else:
-        yield "Invalid model selected.", "Invalid model selected."
-        return
-
-    if video_path is None:
-        yield "Please upload a video.", "Please upload a video."
-        return
-
-    frames = downsample_video(video_path)
-    messages = [
-        {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
-        {"role": "user", "content": [{"type": "text", "text": text}]}
-    ]
-    for frame in frames:
-        image, timestamp = frame
-        messages[1]["content"].append({"type": "text", "text": f"Frame {timestamp}:"})
-        messages[1]["content"].append({"type": "image", "image": image})
-    inputs = processor.apply_chat_template(
-        messages,
-        tokenize=True,
-        add_generation_prompt=True,
-        return_dict=True,
-        return_tensors="pt",
-        truncation=True,
-        max_length=MAX_INPUT_TOKEN_LENGTH
-    ).to(device)
-    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-    generation_kwargs = {
-        **inputs,
-        "streamer": streamer,
-        "max_new_tokens": max_new_tokens,
-        "do_sample": True,
-        "temperature": temperature,
-        "top_p": top_p,
-        "top_k": top_k,
-        "repetition_penalty": repetition_penalty,
-    }
-    thread = Thread(target=model.generate, kwargs=generation_kwargs)
-    thread.start()
-    buffer = ""
-    for new_text in streamer:
-        buffer += new_text
-        buffer = buffer.replace("<|im_end|>", "")
-        time.sleep(0.01)
-        yield buffer, buffer
-
-# Define examples for image and video inference
+# Define examples for image inference
 image_examples = [
     ["Reconstruct the content [table] as it is.", "images/doc.jpg"],
     ["Reconstruct the doc [table] as it is.", "images/zh.png"],
@@ -292,11 +204,6 @@ image_examples = [
     ["OCR the image", "images/image1.png"]
 ]
 
-video_examples = [
-    ["Explain the video in detail", "videos/2.mp4"],
-    ["Explain the video in detail", "videos/1.mp4"]
-]
-
 css = """
 #main-title h1 {
     font-size: 2.3em !important;
@@ -311,17 +218,11 @@ with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
     gr.Markdown("# **core [OCR](https://huggingface.co/collections/prithivMLmods/multimodal-implementations)**", elem_id="main-title")
     with gr.Row():
         with gr.Column(scale=2):
-            with gr.Tabs():
-                with gr.TabItem("Image Inference"):
-                    image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
-                    image_upload = gr.Image(type="pil", label="Upload Image", height=290)
-                    image_submit = gr.Button("Submit", variant="primary")
-                    gr.Examples(examples=image_examples, inputs=[image_query, image_upload])
-                with gr.TabItem("Video Inference"):
-                    video_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
-                    video_upload = gr.Video(label="Upload Video", height=290)
-                    video_submit = gr.Button("Submit", variant="primary")
-                    gr.Examples(examples=video_examples, inputs=[video_query, video_upload])
+            image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
+            image_upload = gr.Image(type="pil", label="Upload Image", height=290)
+            image_submit = gr.Button("Submit", variant="primary")
+            gr.Examples(examples=image_examples, inputs=[image_query, image_upload])
+            
             with gr.Accordion("Advanced options", open=False):
                 max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS)
                 temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6)
@@ -346,11 +247,6 @@ with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
         inputs=[model_choice, image_query, image_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
         outputs=[output, markdown_output]
     )
-    video_submit.click(
-        fn=generate_video,
-        inputs=[model_choice, video_query, video_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
-        outputs=[output, markdown_output]
-    )
 
 if __name__ == "__main__":
     demo.queue(max_size=50).launch(mcp_server=True, ssr_mode=False, show_error=True)