| import os |
| import logging |
| import h5py |
| import numpy as np |
| from tqdm import tqdm |
| from modules import mediapipe_generator |
|
|
| |
| DATA_ROOT = "fsl-data/sentence_data" |
| HDF5_LOCATION = "sentence_landmarks.h5" |
|
|
| logging.basicConfig(level=logging.INFO, format='%(asctime)s [%(levelname)s] %(message)s') |
|
|
| def main(): |
| if not os.path.exists(DATA_ROOT): |
| logging.error(f"Directory {DATA_ROOT} not found!") |
| return |
|
|
| video_files = [f for f in os.listdir(DATA_ROOT) if f.lower().endswith(('.mp4', '.avi', '.mov'))] |
| |
| with h5py.File(HDF5_LOCATION, 'w') as f: |
| for filename in tqdm(video_files, desc="Processing Videos"): |
| video_path = os.path.join(DATA_ROOT, filename) |
| |
| try: |
| data = mediapipe_generator.generate_mediapipe(filepath=video_path) |
| if not data: |
| logging.warning(f"No landmarks detected for {filename}") |
| continue |
|
|
| p_seq = np.array([mediapipe_generator.extract_to_array(r.pose_landmarks, 33, 4) for r in data]) |
| f_seq = np.array([mediapipe_generator.extract_to_array(r.face_landmarks, 468, 3) for r in data]) |
| lh_seq = np.array([mediapipe_generator.extract_to_array(r.left_hand_landmarks, 21, 3) for r in data]) |
| rh_seq = np.array([mediapipe_generator.extract_to_array(r.right_hand_landmarks, 21, 3) for r in data]) |
|
|
| group_key = filename.replace('/', '_') |
| sample_grp = f.create_group(group_key) |
| |
| sample_grp.create_dataset('pose', data=p_seq, compression="gzip") |
| sample_grp.create_dataset('face', data=f_seq, compression="gzip") |
| sample_grp.create_dataset('left_hand', data=lh_seq, compression="gzip") |
| sample_grp.create_dataset('right_hand', data=rh_seq, compression="gzip") |
| |
| sample_grp.attrs['frame_count'] = len(data) |
| sample_grp.attrs['original_filename'] = filename |
|
|
| except Exception as e: |
| logging.error(f"Critical error processing {filename}: {e}") |
|
|
| logging.info(f"HDF5 dataset created successfully at {HDF5_LOCATION}") |
|
|
| if __name__ == "__main__": |
| main() |
|
|
