Update app.py

app.py

@@ -12,7 +12,6 @@ import spaces
 import torch
 import numpy as np
 from PIL import Image, ImageOps
-import cv2
 import requests
 
 from transformers import (
@@ -136,34 +135,14 @@ model_1_5b = AutoModelForImageTextToText.from_pretrained(
     attn_implementation="flash_attention_2"
 ).eval()
 
-
-def downsample_video(video_path):
-    """Downsample a video to evenly spaced frames, returning PIL images with timestamps."""
-    vidcap = cv2.VideoCapture(video_path)
-    total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
-    fps = vidcap.get(cv2.CAP_PROP_FPS)
-    frames = []
-    # Use a smaller number of frames for video to avoid overwhelming the model
-    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(model_name: str, text: str, media_input, media_type: str,
-             max_new_tokens: int = 1024,
-             temperature: float = 0.6,
-             top_p: float = 0.9,
-             top_k: int = 50,
-             repetition_penalty: float = 1.2):
-    """Generic generation function for both image and video."""
+def generate_image(model_name: str, text: str, image: Image.Image,
+                   max_new_tokens: int = 1024,
+                   temperature: float = 0.6,
+                   top_p: float = 0.9,
+                   top_k: int = 50,
+                   repetition_penalty: float = 1.2):
+    """Generation function for image input."""
     if model_name == "Nanonets-OCR2-3B":
         processor, model = processor_3b, model_3b
     elif model_name == "Nanonets-OCR2-1.5B-exp":
@@ -172,30 +151,20 @@ def generate(model_name: str, text: str, media_input, media_type: str,
         yield "Invalid model selected.", "Invalid model selected."
         return
 
-    if media_input is None:
-        yield f"Please upload an {media_type}.", f"Please upload an {media_type}."
+    if image is None:
+        yield "Please upload an image.", "Please upload an image."
         return
         
-    if media_type == "image":
-        images = [media_input]
-    elif media_type == "video":
-        frames = downsample_video(media_input)
-        images = [frame for frame, _ in frames]
-    else:
-        yield "Invalid media type.", "Invalid media type."
-        return
+    images = [image]
 
     messages = [
         {
             "role": "user",
-            "content": [{"type": "image"} for _ in images] + [
-                {"type": "text", "text": text}
-            ]
+            "content": [{"type": "image"}] + [{"type": "text", "text": text}]
         }
     ]
     
     prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
-    # Since device_map="auto" is used, we don't need .to(device)
     inputs = processor(text=prompt, images=images, return_tensors="pt")
 
     streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
@@ -216,15 +185,7 @@ def generate(model_name: str, text: str, media_input, media_type: str,
         buffer += new_text.replace("<|im_end|>", "")
         yield buffer, buffer
 
-# Wrapper functions for Gradio clarity
-def generate_image(*args):
-    yield from generate(*args[:3], media_input=args[2], media_type="image", *args[3:])
-
-def generate_video(*args):
-    yield from generate(*args[:3], media_input=args[2], media_type="video", *args[3:])
-
-
-# Define examples for image and video inference
+# Define examples for image inference
 image_examples = [
     ["Reconstruct the doc [table] as it is.", "images/0.png"],
     ["Describe the image!", "images/8.png"],
@@ -237,27 +198,17 @@ image_examples = [
     ["Convert formula to late.", "images/7.jpg"],
 ]
 
-video_examples = [
-    ["Explain the video in detail.", "videos/1.mp4"],
-    ["Explain the video in detail.", "videos/2.mp4"]
-]
-
 # Create the Gradio Interface
 with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
     gr.Markdown("# **Multimodal OCR3**", 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 (<= 30s)", height=290)
-                    video_submit = gr.Button("Submit", variant="primary")
-                    gr.Examples(examples=video_examples, inputs=[video_query, video_upload])
+            # Image Inference Components
+            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)
@@ -282,11 +233,6 @@ 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=[raw_output, formatted_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=[raw_output, formatted_output]
-    )
 
 if __name__ == "__main__":
     demo.queue(max_size=50).launch(ssr_mode=False, show_error=True)