app.py

from transformers import BitsAndBytesConfig, LlavaNextVideoForConditionalGeneration, LlavaNextVideoProcessor
import torch
import numpy as np
import av
import spaces
import gradio as gr
import os
import json

# Model Configuration
quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16
)

model_name = 'llava-hf/LLaVA-NeXT-Video-7B-DPO-hf'  

# Load Model and Processor
processor = LlavaNextVideoProcessor.from_pretrained(model_name)
model = LlavaNextVideoForConditionalGeneration.from_pretrained(
    model_name,
    quantization_config=quantization_config,
    device_map='auto'
)

@spaces.GPU
def read_video_pyav(container, indices):
    '''
    Decode the video with PyAV decoder.
    '''
    frames = []
    container.seek(0)
    start_index = indices[0]
    end_index = indices[-1]
    for i, frame in enumerate(container.decode(video=0)):
        if i > end_index:
            break
        if i >= start_index and i in indices:
            frames.append(frame)
    return np.stack([x.to_ndarray(format="rgb24") for x in frames])

@spaces.GPU
def process_video(video_file, question):
    '''
    Processes a single video and returns the answer to the given question.
    '''
    with av.open(video_file.name) as container:
        total_frames = container.streams.video[0].frames
        indices = np.arange(0, total_frames, total_frames / 8).astype(int)
        video_clip = read_video_pyav(container, indices)

    conversation = [
        {
            "role": "user",
            "content": [
                {"type": "text", "text": f"{question}"},
                {"type": "video"},
            ],
        },
    ]
    prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)
    input = processor([prompt], videos=[video_clip], padding=True, return_tensors="pt").to(model.device)
    generate_kwargs = {"max_new_tokens": 100, "do_sample": True, "top_p": 0.9}
    output = model.generate(**input, **generate_kwargs)
    generated_text = processor.batch_decode(output, skip_special_tokens=True)[0]
    return generated_text.split("ASSISTANT: ", 1)[-1].strip()

@spaces.GPU
def analyze_videos(video_files, selected_questions):
    """Analyzes all videos with the selected questions."""
    all_results = {}

    questions = {
        "hands_free": "Examine the subject’s right and left hands in the video to check if they are holding anything like a microphone, book, paper(White color), object, or any electronic device, try segmentations and decide if the hands are free or not.",
        "standing/sitting": "Evaluate the subject’s body posture and movement within the video. Are they standing upright with both feet planted firmly on the ground? If so, they are standing. If they seem to be seated, they are seated.",
        "interaction_with_background": "Assess the surroundings behind the subject in the video. Do they seem to interact with any visible screens, such as laptops, TVs, or digital billboards? If yes, then they are interacting with a screen. If not, they are not interacting with a screen.",
        "indoors/outdoors": "Consider the broader environmental context shown in the video’s background. Are there signs of an open-air space, like greenery, structures, or people passing by? If so, it’s an outdoor setting. If the setting looks confined with furniture, walls, or home decorations, it’s an indoor environment."
    }

    for video_file in video_files:
        video_name = os.path.basename(video_file.name)
        all_results[video_name] = {}
        for question_key in selected_questions:
            answer = process_video(video_file, questions[question_key])
            # Simple True/False determination (You might want to refine this)
            all_results[video_name][question_key] = "true" if "yes" in answer.lower() else "false"
    return json.dumps(all_results, indent=4)

# Define Gradio interface 
iface = gr.Interface(
    fn=analyze_videos,
    inputs=[
        gr.File(label="Upload Videos", file_count="multiple"),
        gr.CheckboxGroup(["hands_free", "standing/sitting", "interaction_with_background", "indoors/outdoors"], 
                        label="Select Questions to Apply") 
    ],
    outputs=gr.JSON(label="Analysis Results"), 
    title="Video Analysis",
    description="Upload videos and select questions to analyze."
)

if __name__ == "__main__":
    iface.launch(debug=True)