Download validate_release.py from diffracting/egocentric-maintenance-sample: direct link, hf CLI and curl.
- Browser
- Download file 5.39 kB
-
https://huggingface.co/datasets/diffracting/egocentric-maintenance-sample/resolve/main/validate_release.py
- Command line
-
hf download hf://datasets/diffracting/egocentric-maintenance-sample/validate_release.py
-
curl -L -o validate_release.py https://huggingface.co/datasets/diffracting/egocentric-maintenance-sample/resolve/main/validate_release.py
5.39 kB
| """Validate a portable RGB-only release without the processing stack.""" | |
| from pathlib import Path | |
| import sys,json,hashlib,argparse,subprocess | |
| sys.dont_write_bytecode=True | |
| import cv2 | |
| import numpy as np | |
| import polars as pl | |
| def validate(root): | |
| root=Path(root).resolve();sys.path.insert(0,str(root)) | |
| from observation_reader import ObservationDataset,camera_points | |
| import pyarrow.parquet as pq | |
| ds=ObservationDataset(root) | |
| assert ds.info['capture_mode']=='rgb_only' and ds.info['license']=='cc-by-4.0' | |
| assert ds.episodes | |
| report={'status':'passed','scope':'Full video decode, displayed PTS, frame identity tables, missing modalities, segment bounds, embedded preview schema and all checksums. No annotation accuracy or model training evaluation.','episodes':[]} | |
| for key,e in ds.episodes.items(): | |
| mapping=ds.table(key,'frame_mapping');n=e['frame_count'] | |
| assert mapping.height==n and mapping['rgb_frame_index'].to_list()==list(range(n)) | |
| source=mapping['source_frame_index'].to_numpy() | |
| assert np.all(np.diff(source)==1) and source[0]==e['source']['selection']['first_source_frame'] and source[-1]==e['source']['selection']['last_source_frame'] | |
| ts=mapping['t_s'].to_numpy();assert ts[0]==0 and np.all(np.isfinite(ts)) and np.all(np.diff(ts)>0) | |
| assert np.max(np.abs(np.diff(mapping['source_pts_s'].to_numpy())-np.diff(ts)))<1e-6 | |
| assert mapping['rgb_decodable'].all() | |
| probe=json.loads(subprocess.check_output(['ffprobe','-v','error','-show_streams','-show_packets','-of','json',str(root/e['video'])])) | |
| assert len(probe['streams'])==1 | |
| stream=probe['streams'][0] | |
| assert stream['codec_name']=='h264' and stream['pix_fmt']=='yuv420p' and stream['color_transfer']=='bt709' | |
| pts=sorted(float(p['pts_time']) for p in probe['packets'] if 'pts_time' in p and 'D' not in p.get('flags','')) | |
| assert len(pts)==n and np.max(np.abs(np.array(pts)-mapping['video_pts_s'].to_numpy()))<2e-6 | |
| cap=cv2.VideoCapture(str(root/e['video']));decoded=0 | |
| try: | |
| while True: | |
| ok,frame=cap.read() | |
| if not ok:break | |
| assert frame.shape==(e['height'],e['width'],3) | |
| decoded+=1 | |
| finally:cap.release() | |
| assert decoded==n,(key,decoded,n) | |
| indices=sorted({0,n//4,n//2,3*n//4,n-1}) | |
| for index in indices: | |
| frame=ds.frame(key,index) | |
| assert frame['source_frame_index']==int(source[index]) and frame['timestamp_s']==ts[index] | |
| assert all(frame[k] is None for k in ('depth_m','depth_valid','confidence','K_rgb','camera_pose_xyzw','sensor_timestamp_s')) | |
| assert not any(v for k,v in frame['modalities'].items() if k!='rgb') | |
| try:camera_points(frame) | |
| except ValueError:pass | |
| else:raise AssertionError('RGB-only geometry must be unavailable') | |
| for name in ('hand_pose_raw','hand_pose'): | |
| if name not in e['signals']:continue | |
| hand=ds.table(key,name) | |
| assert all(hand[k].null_count()==hand.height for k in ('z','wx','wy','wz')) | |
| assert ds.annotations_near(key,name,1e6).height==0 | |
| assert hashlib.file_digest((root/e['video']).open('rb'),'sha256').hexdigest()==e['video_sha256'] | |
| report['episodes'].append({'capture_id':key,'decoded_frames':decoded,'timestamp_rows':n,'random_access_indices':indices,'metric_channels_absent':True}) | |
| print(key,decoded,'decoded frames and timestamps verified',flush=True) | |
| segments=pl.read_parquet(root/'annotations/action_segments.parquet') | |
| for row in segments.to_dicts(): | |
| assert row['capture_id'] in ds.episodes | |
| assert 0<=row['start_s']<row['end_s']<=ds.episodes[row['capture_id']]['duration_s'] | |
| assert row['annotation_source']=='assistant_visual_review' and row['boundary_precision_s']==2.0 | |
| catalog=pq.read_table(root/'data/episodes.parquet') | |
| features=json.loads(catalog.schema.metadata[b'huggingface'])['info']['features'] | |
| assert features['video']['_type']=='Video' and catalog.num_rows==len(ds.episodes) | |
| for row in catalog.to_pylist(): | |
| assert row['video']['bytes']==(root/ds.episodes[row['capture_id']]['preview']).read_bytes() | |
| seen=set() | |
| for line in (root/'checksums.sha256').read_text().splitlines(): | |
| expected,rel=line.split(' ',1);target=(root/rel).resolve() | |
| assert target.is_relative_to(root) and rel not in seen | |
| assert hashlib.file_digest(target.open('rb'),'sha256').hexdigest()==expected,rel | |
| seen.add(rel) | |
| actual={p.relative_to(root).as_posix() for p in root.rglob('*') if p.is_file() and p.name!='checksums.sha256' and '__pycache__' not in p.parts and '.cache' not in p.parts} | |
| assert actual==seen,(actual-seen,seen-actual) | |
| report.update(verified_artifact_hashes=len(seen),action_segments=segments.height,huggingface_video_rows=catalog.num_rows,total_decoded_frames=sum(e['decoded_frames'] for e in report['episodes'])) | |
| return report | |
| if __name__=='__main__': | |
| p=argparse.ArgumentParser(description=__doc__);p.add_argument('root',type=Path);p.add_argument('--write-report',action='store_true');args=p.parse_args() | |
| report=validate(args.root) | |
| if args.write_report:(args.root/'validation.json').write_text(json.dumps(report,indent=2),encoding='utf-8') | |
| print(json.dumps(report,indent=2)) | |