import gradio as gr
from transformers import pipeline
import cv2

def video_to_descriptions(video):
    # Charger le modèle via pipeline
    model = pipeline('image-to-text', model='nlpconnect/vit-gpt2-image-captioning')
    
    # Ouvrir la vidéo
    cap = cv2.VideoCapture(video.name)
    fps = int(cap.get(cv2.CAP_PROP_FPS))
    
    descriptions = []
    frame_count = 0
    
    while True:
        ret, frame = cap.read()
        if not ret:
            break
        
        # Extraire une image toutes les demi-secondes
        if frame_count % (fps // 2) == 0:
            # Convertir l'image en RGB
            frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            # Obtenir la description de l'image
            outputs = model(frame_rgb)
            description = outputs[0]['describe-text']
            descriptions.append(description)
        
        frame_count += 1

    # Fermer le lecteur vidéo
    cap.release()

    # Concaténer les descriptions
    concatenated_descriptions = " ".join(descriptions)
    
    return concatenated_descriptions

iface = gr.Interface(
    fn=video_to_descriptions,
    inputs=gr.inputs.Video(type="file", label="Importez une vidéo"),
    outputs="text",
    live=False
)

if __name__ == "__app__":
    iface.launch()