Download code/transfer/collect_all_old256_inventory.py from FrontRear/VideoObjectMover: direct link, hf CLI and curl.
- Browser
- Download file 2.19 kB
-
https://huggingface.co/datasets/FrontRear/VideoObjectMover/resolve/main/code/transfer/collect_all_old256_inventory.py
- Command line
-
hf download hf://datasets/FrontRear/VideoObjectMover/code/transfer/collect_all_old256_inventory.py
-
curl -L -o collect_all_old256_inventory.py https://huggingface.co/datasets/FrontRear/VideoObjectMover/resolve/main/code/transfer/collect_all_old256_inventory.py
2.19 kB
| """Remote read-only inventory for the old 256-pair release. | |
| Run via `ssh 6000D ... python -B - < this_script > evidence/file.json`. | |
| It does not edit the training machine, and does not claim an index is usable data. | |
| """ | |
| import json | |
| import pathlib | |
| import re | |
| import sys | |
| base = pathlib.Path('/root/autodl-tmp/LampEditTraining/data_v2') | |
| bbox = base/'bbox_conditions_v003_20261021' | |
| index = json.loads((bbox/'INDEX.json').read_text()) | |
| sources = [base/'train_manifest.mixed256.json', base/'train_manifest.new118.json'] | |
| groups = index['groups'] | |
| sources.extend(pathlib.Path(g['manifest']) for g in groups) | |
| pat_data = re.compile(rb'data_v2/assets/([0-9a-f]{32})') | |
| pat_bbox = re.compile(rb'bbox_v3/assets/([0-9a-f]{32})') | |
| data_ids, bbox_ids = set(), set() | |
| missing = [] | |
| for p in sources: | |
| if not p.is_file(): | |
| missing.append(str(p)) | |
| continue | |
| with p.open('rb') as f: | |
| for line in f: | |
| data_ids.update(x.decode('ascii') for x in pat_data.findall(line)) | |
| bbox_ids.update(x.decode('ascii') for x in pat_bbox.findall(line)) | |
| # Each factored render can itself point to base data assets and bbox assets. | |
| recipes = 0 | |
| for asset in sorted(bbox_ids): | |
| path = bbox/'assets'/asset | |
| if not path.is_dir(): | |
| missing.append(str(path)) | |
| continue | |
| for recipe in path.rglob('RENDER.json'): | |
| b = recipe.read_bytes() | |
| data_ids.update(x.decode('ascii') for x in pat_data.findall(b)) | |
| bbox_ids.update(x.decode('ascii') for x in pat_bbox.findall(b)) | |
| recipes += 1 | |
| output = { | |
| 'scope': 'old 256 manifest and indexed bbox resources; additional dependencies may exist', | |
| 'index_groups': len(groups), 'pair_count_in_bbox_index': index.get('pairs'), | |
| 'mixed_and_new118_manifests': [str(p) for p in sources[:2]], | |
| 'data_asset_ids': sorted(data_ids), 'bbox_asset_ids': sorted(bbox_ids), | |
| 'bbox_group_paths': sorted(str(pathlib.Path(g['manifest']).parent.relative_to(base)) for g in groups), | |
| 'factored_render_recipes_scanned': recipes, 'missing_source_paths': missing, | |
| 'no_512_data_included_by_design': True, 'remote_mutations': False, | |
| } | |
| json.dump(output, sys.stdout, ensure_ascii=False) | |
| sys.stdout.write('\n') | |