app.py

import gradio as gr
import os
from inference import predict_activity

# Configure these paths
DATASET = "ucf11"  # or 'ucf50'
MODEL_PATH = "models/ucf11_lstm_model.pt"  # update if different


def predict_from_file(video_file):
    # video_file is a path returned by Gradio when uploading
    if video_file is None:
        return "No file uploaded"
    try:
        cls_idx, label = predict_activity(DATASET, video_file, MODEL_PATH)
        return f"Predicted: {cls_idx} — {label}"
    except Exception as e:
        return f"Error: {e}"


with gr.Blocks() as demo:
    gr.Markdown("# Human Activity Recognition - Video Inference")
    with gr.Row():
        video_in = gr.Video(label="Upload video")
        out = gr.Textbox(label="Prediction")
    btn = gr.Button("Predict")
    btn.click(fn=predict_from_file, inputs=[video_in], outputs=[out])


if __name__ == "__main__":
    # On Hugging Face Spaces, running `python app.py` should start Gradio
    demo.launch(server_name="0.0.0.0", server_port=7860)