app.py

import os
import subprocess
import importlib.util
import gradio as gr
import logging
#from moviepy.editor import VideoFileClip
import torch
#import spaces

ACCESS_KEY = os.getenv("ACCESS_KEY")

'''
torch.use_deterministic_algorithms(True)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.backends.cuda.matmul.allow_tf32 = False
torch.backends.cudnn.allow_tf32 = False


def truncate_video(video_file):
    clip = VideoFileClip(video_file)
    truncated_clip = clip.subclip(0, min(15, clip.duration))
    truncated_video_file = "temp_truncated_video.mp4"
    truncated_clip.write_videofile(truncated_video_file, codec="libx264", audio_codec="aac")
    return truncated_video_file
'''
def clone_repo():
    repo_url = "https://github.com/NeeravSood/AllMark-MVP.git"
    repo_path = "./repository"

    github_pat = os.getenv("GITHUB_PAT")
    if not github_pat:
        raise RuntimeError("GitHub Personal Access Token (GITHUB_PAT) not found in environment variables.")
    authenticated_repo_url = f"https://{github_pat}@github.com/NeeravSood/AllMark-MVP.git"

    if os.path.exists(repo_path):
        print("Repository already cloned.")
    else:
        try:
            subprocess.run(
                ["git", "clone", authenticated_repo_url, repo_path],
                check=True,
                text=True,
                capture_output=True
            )
            print("Repository cloned successfully.")
        except subprocess.CalledProcessError as e:
            print("Output:", e.stdout)
            print("Error:", e.stderr)
            raise RuntimeError(f"Failed to clone repository: {e.stderr}")

def import_backend_script(script_name):
    """Dynamically import the backend script."""
    try:
        script_path = os.path.join("./repository", script_name)
        if not os.path.exists(script_path):
            raise FileNotFoundError(f"Script {script_name} not found in the repository.")
        
        spec = importlib.util.spec_from_file_location("backend_module", script_path)
        backend_module = importlib.util.module_from_spec(spec)
        spec.loader.exec_module(backend_module)
        return backend_module
    except Exception as e:
        logging.error(f"Error importing backend script: {str(e)}")
        raise RuntimeError(f"Failed to import backend script: {str(e)}")

# Run repository setup and model import
clone_repo()
backend = import_backend_script("app.py") 
analyzer = backend.DeepfakeAnalyzer()

#@spaces.GPU(duration=3000)
def analyze_video(video_file):
    if ACCESS_KEY is None:
        logging.error("Access key not set in environment variables.")
        return {"error": "Server misconfiguration. Access key not set."}

    try:
        #truncated_video = truncate_video(video_file)  # Truncate the video for faster analysis
        #results = analyzer.analyze_media(truncated_video)
        results = analyzer.analyze_media(video_file)


        # Extract frame count and video probability
        frame_count = len(results['video_analysis']['frame_results'])
        video_probability = results['video_analysis']['probability']
        combined_assessment = results.get('combined_assessment', "Inconclusive")

        # Reapply combined_results logic for front-end clarity
        if frame_count < 300:
            assessment = "Inconclusive due to lack of sufficient frames"
        elif frame_count < 500:
            threshold = 0.4
            assessment = "Deepfake" if video_probability >= threshold else "Genuine"
        else:
            threshold = 0.5
            assessment = "Deepfake" if video_probability >= threshold else "Genuine"

        # Construct output message
        message = (
            f"According to our analysis, the video you uploaded appears to be {assessment}. "
            f"{frame_count} frames were analyzed in total."
        )

        # Return the final output
        output = {
            "message": message,
            "details": {
                "video_probability": video_probability,
                "combined_assessment": assessment  # Final assessment based on reapplied logic
            }
        }
        return output

    except Exception as e:
        logging.error(f"Error during analysis: {e}")
        return {"error": "An error occurred during video analysis. Please check your input and try again."}




interface = gr.Interface(
    fn=analyze_video,
    inputs=gr.Video(label="Upload Video"),
    outputs="json",
    title="AllMark - Deepfake and AI Video Analyzer",
    description="Upload a MP4 video for analysis. Multiple model iterations are being tested, so results are slow and may vary for the same video. Our Models 2 and 3 shall be available on AWS."
)

if __name__ == "__main__":
    interface.launch()