| """ |
| Merge metadata from all GPUs after multi-GPU preprocessing. |
| |
| Run this after all GPUs have completed processing. |
| """ |
|
|
| import json |
| from pathlib import Path |
| import argparse |
|
|
|
|
| def main(): |
| parser = argparse.ArgumentParser() |
| parser.add_argument('--output-dir', type=str, required=True, help='Output directory') |
| parser.add_argument('--num-gpus', type=int, default=16, help='Total number of GPUs') |
| args = parser.parse_args() |
|
|
| output_dir = Path(args.output_dir) |
|
|
| print("=" * 80) |
| print("Merging Multi-GPU Metadata") |
| print("=" * 80) |
|
|
| all_metadata = [] |
| total_processed = 0 |
| total_skipped = 0 |
| total_previews = 0 |
| missing_gpus = [] |
|
|
| for gpu_id in range(args.num_gpus): |
| metadata_path = output_dir / f'metadata_gpu{gpu_id:02d}.json' |
|
|
| if not metadata_path.exists(): |
| print(f"⚠️ Missing: GPU {gpu_id} (file not found)") |
| missing_gpus.append(gpu_id) |
| continue |
|
|
| with open(metadata_path, 'r') as f: |
| metadata = json.load(f) |
| all_metadata.append(metadata) |
| total_processed += metadata['processed_episodes'] |
| total_skipped += metadata['skipped_episodes'] |
| total_previews += metadata['preview_videos'] |
|
|
| print(f"✓ GPU {gpu_id:2d}: {metadata['processed_episodes']:5d} processed, " |
| f"{metadata['skipped_episodes']:4d} skipped, " |
| f"{metadata['preview_videos']:2d} previews") |
|
|
| print("\n" + "=" * 80) |
| print("Summary") |
| print("=" * 80) |
| print(f" GPUs completed: {len(all_metadata)}/{args.num_gpus}") |
| print(f" Total processed: {total_processed}") |
| print(f" Total skipped: {total_skipped}") |
| print(f" Total previews: {total_previews}") |
|
|
| if missing_gpus: |
| print(f"\n⚠️ Missing GPUs: {missing_gpus}") |
| print(" Some GPUs may still be running. Check logs:") |
| for gpu_id in missing_gpus: |
| machine_id = gpu_id // 8 |
| local_gpu = gpu_id % 8 |
| log_file = output_dir / f'log_machine{machine_id}_gpu{local_gpu}.txt' |
| if log_file.exists(): |
| print(f" tail -f {log_file}") |
|
|
| |
| merged_metadata = { |
| 'num_gpus_completed': len(all_metadata), |
| 'num_gpus_total': args.num_gpus, |
| 'total_processed_episodes': total_processed, |
| 'total_skipped_episodes': total_skipped, |
| 'total_preview_videos': total_previews, |
| 'missing_gpus': missing_gpus, |
| 'gpu_metadata': all_metadata, |
| 'camera_params': all_metadata[0]['camera_params'] if all_metadata else {}, |
| 'point_distribution': all_metadata[0]['point_distribution'] if all_metadata else {}, |
| 'image_resolution': all_metadata[0]['image_resolution'] if all_metadata else [], |
| 'max_frames_per_episode': all_metadata[0]['max_frames_per_episode'] if all_metadata else 400, |
| 'cotracker_model': all_metadata[0]['cotracker_model'] if all_metadata else '', |
| 'cotracker_chunking': all_metadata[0]['cotracker_chunking'] if all_metadata else '', |
| } |
|
|
| merged_path = output_dir / 'metadata_merged.json' |
| with open(merged_path, 'w') as f: |
| json.dump(merged_metadata, f, indent=2) |
|
|
| print(f"\n✓ Merged metadata saved: {merged_path}") |
| print("=" * 80) |
|
|
| |
| data_dir = output_dir / 'data' |
| if data_dir.exists(): |
| npz_files = list(data_dir.glob('episode_*.npz')) |
| print(f"\nNPZ files in data/: {len(npz_files)}") |
|
|
| |
| preview_dir = output_dir / 'preview_videos' |
| if preview_dir.exists(): |
| video_files = list(preview_dir.glob('preview_episode_*.mp4')) |
| print(f"Preview videos: {len(video_files)}") |
|
|
| print("\nProcessing complete! ✓") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
