import gradio as gr
from predictor import predict_video_with_cam

def inference(video):
    if video is None:
        return "Please upload or select a video.", []

    cams, final_label, final_conf, faces_detected, total_frames = predict_video_with_cam(video, max_frames=5)

    if faces_detected == 0:
        return "No face detected in any of the frames. Please try a different video.", []

    summary = f"Video-Level Prediction → {final_label} ({final_conf:.2%})"
    frame_lines = [f"Frame {i+1} → {label} ({conf:.2%})" for i, (label, conf, _) in enumerate(cams)]
    result_text = summary + "\n" + "\n".join(frame_lines)

    if faces_detected < total_frames:
        result_text += f"\nOnly {faces_detected} out of {total_frames} frames contained detectable faces."

    images = [img for _, _, img in cams]
    return result_text, images

with gr.Blocks() as demo:
    # Title and description (centered with HTML)
    gr.Markdown(
        """
        <div style="text-align: center;">
            <h2>Deepfake Detection App with Grad-CAM</h2>
            <p>Upload your own video or choose one of the examples below to see how the model works.</p>
            <p><strong>Model Info:</strong> EfficientNet-B0 trained on a FaceForensics++ subset. Grad-CAM is used to improve explainability.</p>
        </div>
        """
    )

    with gr.Row():
        video_input = gr.Video(label="Upload or select a video")
        prediction_output = gr.Textbox(label="Prediction Summary", lines=8, interactive=False)

    gradcam_output = gr.Gallery(
        label="Grad-CAM Visuals",
        columns=5,
        height="200px",
        object_fit="scale-down",
        show_label=True,
        container=True
    )

    with gr.Row():
        predict_btn = gr.Button("Predict")
        clear_btn = gr.Button("Clear")

    predict_btn.click(fn=inference, inputs=video_input, outputs=[prediction_output, gradcam_output])
    clear_btn.click(fn=lambda: (None, "", []), inputs=[], outputs=[video_input, prediction_output, gradcam_output])

    # Example videos (smaller and labeled)
    gr.Markdown("### Try with Example Videos")

    example_videos = [
        ("videos/fake1.mp4", "Fake Video 1"),
        ("videos/fake2.mp4", "Fake Video 2"),
        ("videos/fake3.mp4", "Fake Video 3"),
        ("videos/real1.mp4", "Real Video 1"),
        ("videos/real2.mp4", "Real Video 2"),
        ("videos/real3.mp4", "Real Video 3"),
        ("videos/no_face.mp4", "No Face Video"),
    ]

    with gr.Row():
        for path, label in example_videos:
            with gr.Column(scale=1, min_width=150):
                gr.Markdown(f"<center><sub>{label}</sub></center>")
                example_video = gr.Video(value=path, interactive=False, height=100, show_label=False)
                example_button = gr.Button("Use this video")
                example_button.click(fn=lambda p=path: p, inputs=[], outputs=video_input)

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