import pandas as pd
import subprocess
import os
from tqdm import tqdm

def slice_sign_dataset(csv_path, video_dir="fsl-data/raw_videos", output_dir="fsl-data/sliced_dataset"):
    df = pd.read_csv(csv_path)
    
    print(f"[*] Beginning slicing of {len(df)} samples...")
    
    for idx, row in tqdm(df.iterrows(), total=len(df)):
        filename = row['filename']
        label = row['label']
        start_f = int(row['start_frame'])
        end_f = int(row['end_frame'])
        
        video_stem = os.path.splitext(filename)[0]
        target_folder = os.path.join(output_dir, video_stem, label)
        os.makedirs(target_folder, exist_ok=True)
        
        output_filename = f"sample_{idx:04d}.mp4"
        output_path = os.path.join(target_folder, output_filename)
        
        if os.path.exists(output_path):
            continue
            
        start_sec = start_f / 30.0
        duration_sec = (end_f - start_f) / 30.0
        
        input_path = os.path.join(video_dir, filename)
        
        cmd = [
            'ffmpeg',
            '-y',
            '-ss', str(start_sec),
            '-i', input_path,
            '-t', str(duration_sec),
            '-c', 'copy',
            '-loglevel', 'error',
            output_path
        ]
        
        try:
            subprocess.run(cmd, check=True)
        except subprocess.CalledProcessError as e:
            print(f"[!] Error slicing {filename} at index {idx}: {e}")

if __name__ == "__main__":
    slice_sign_dataset("master_annotations.csv")