File size: 4,422 Bytes
3f3fbe2 3726258 3f3fbe2 3726258 3f3fbe2 3726258 3f3fbe2 3726258 3f3fbe2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 | #!/usr/bin/env python3
"""Write ``<tier>/annotations_complete.json`` (the scenes in which every intended class is annotated) and a
CSV of completeness per sub-tier.
"""
from __future__ import annotations
import argparse
import ast
import csv
import datetime
import json
import pathlib
import sys
from typing import Dict, List, Optional
sys.path.insert(0, str(pathlib.Path(__file__).resolve().parents[2]))
from strangebench.common.variants import V1_LABELS, order_tiers # noqa: E402
TIERS = list(V1_LABELS)
def discover_tiers(root: pathlib.Path) -> List[str]:
return order_tiers([p.parent.name for p in root.glob("*/annotations.json")])
def complete_subset(d: Dict, anchor_of: Optional[Dict[str, str]] = None) -> Dict:
"""Return ``{"coco": filtered, "n_images", "n_complete", "n_anchor_unannotated"}`` for
one tier's annotations.json. The anchor per scene comes from ``anchor_of`` (scene_specs),
else the first intended label."""
anchor_of = anchor_of or {}
cat = {c['id']: c['name'].lower() for c in d['categories']}
present: Dict[int, set] = {}
for an in d['annotations']:
present.setdefault(an['image_id'], set()).add(cat[an['category_id']])
keep, n_missing_anchor = [], 0
for im in d['images']:
li = im.get('labels_intended', '[]')
intended = {x.lower() for x in (json.loads(li.replace("'", '"')) if isinstance(li, str) else li)}
ok = bool(intended) and intended <= present.get(im['id'], set())
if ok:
keep.append(im['id'])
anchor = anchor_of.get(im.get('scene_id')) or (next(iter(intended)) if intended else None)
if anchor and anchor not in present.get(im['id'], set()):
n_missing_anchor += 1
ks = set(keep)
out = dict(d)
out['images'] = [im for im in d['images'] if im['id'] in ks]
out['annotations'] = [an for an in d['annotations'] if an['image_id'] in ks]
out.setdefault('info', {})
out['info'] = dict(out['info'], completeness_filter='every intended class annotated')
return {"coco": out, "n_images": len(d['images']), "n_complete": len(keep),
"n_anchor_unannotated": n_missing_anchor}
def main() -> None:
ap = argparse.ArgumentParser(); ap.add_argument('--root', default='data/strange_bench/clean')
ap.add_argument('--out-csv', default='strange_bench_annotation_completeness.csv')
ap.add_argument('--scene-specs', default='scene_specs.json', help='anchor name per scene_id (authoritative)')
a = ap.parse_args()
_specs = json.load(open(a.scene_specs)) if pathlib.Path(a.scene_specs).exists() else []
anchor_of = {s_['scene_id']: ast.literal_eval(s_['anchor'])['name'].lower() if isinstance(s_['anchor'], str) else s_['anchor']['name'].lower() for s_ in _specs}
root = pathlib.Path(a.root)
rows = []
for t in discover_tiers(root):
p = root / t / 'annotations.json'
d = json.load(open(p))
res = complete_subset(d, anchor_of)
out = res["coco"]
out['info'] = dict(out['info'], source=str(p), date=str(datetime.date.today()))
json.dump(out, open(p.with_name('annotations_complete.json'), 'w'))
n = res["n_images"]
rows.append(dict(tier=t, n_images=n, n_complete=res["n_complete"], retention_pct=round(100*res["n_complete"]/max(n, 1), 1),
n_anchor_unannotated=res["n_anchor_unannotated"], anchor_unannotated_pct=round(100*res["n_anchor_unannotated"]/max(n, 1), 1)))
print(rows[-1])
if not rows:
raise SystemExit(f"no <tier>/annotations.json under {root}")
with open(a.out_csv, 'w') as f:
w = csv.DictWriter(f, fieldnames=list(rows[0].keys())); w.writeheader(); w.writerows(rows)
tot = sum(r['n_images'] for r in rows); comp = sum(r['n_complete'] for r in rows); ma = sum(r['n_anchor_unannotated'] for r in rows)
f.write(f"# description: per-tier annotation completeness of the released clean STRANGE-Bench tree; annotations_complete.json written next to each annotations.json\n")
f.write(f"# totals: images={tot} complete={comp} ({100*comp/tot:.1f}%) anchor_unannotated={ma} ({100*ma/tot:.1f}%); anchor from scene_specs.json (fallback: first intended label)\n")
f.write(f"# root: {a.root} | script: strangebench/audit/make_complete_annotations.py | generated: {datetime.date.today()}\n")
print('wrote', a.out_csv)
if __name__ == "__main__":
main()
|