| |
| """Standalone sanity check for exact-match *_exact_processed dirs. |
| |
| Verifies, per dataset/split: |
| - every STORED template candidate has a class-set EQUAL to the sample's classes (exact), |
| - no sample matches itself (self-exclusion preserved), |
| - exact-match coverage among foreground-bearing samples (how many got >=1 template), |
| and prints PASS/FAIL. |
| |
| Usage: |
| python sanity_check_exact.py \ |
| ACDC_1percent_exact_processed ACDC_5percent_exact_processed \ |
| ACDC_10percent_exact_processed ACDC_20percent_exact_processed \ |
| MMs_10percent_exact_processed |
| (no args -> checks the default set above) |
| """ |
| import json, os, glob, sys |
| import numpy as np |
| from PIL import Image |
|
|
| RL = "/projects/p32685/RL_dataset" |
| DEFAULT = [ |
| "ACDC_1percent_exact_processed", "ACDC_5percent_exact_processed", |
| "ACDC_10percent_exact_processed", "ACDC_20percent_exact_processed", |
| "MMs_10percent_exact_processed", |
| ] |
|
|
|
|
| def bank_sets(proc): |
| out = {} |
| for npz in glob.glob(os.path.join(proc, "template", "*.npz")): |
| tid = os.path.splitext(os.path.basename(npz))[0] |
| d = np.load(npz, allow_pickle=True) |
| out[tid] = frozenset(int(c) for c in d.get("classes", [])) |
| return out |
|
|
|
|
| def mask_set(p): |
| for c in [p, "/projects/p32685/" + p.lstrip("/")]: |
| if os.path.exists(c): |
| m = np.array(Image.open(c)) |
| return frozenset(int(v) for v in np.unique(m) if v != 0) |
| return None |
|
|
|
|
| def check(name): |
| proc = os.path.join(RL, name) |
| if not os.path.isdir(proc): |
| print(f" {name}: MISSING"); return False |
| bank = bank_sets(proc) |
| all_pass = True |
| for split in ["val_entries.json", "train_entries.json"]: |
| path = os.path.join(proc, "backup_entries", split) |
| if not os.path.exists(path): |
| continue |
| ent = json.load(open(path)) |
| n = with_t = ok = bad = selfm = 0 |
| for e in ent: |
| sc = mask_set(e["label_mask"]) |
| if sc is None or len(sc) == 0: |
| continue |
| n += 1 |
| ss = os.path.splitext(os.path.basename(e["image"]))[0] |
| tpl = e.get("template") |
| tpl = json.loads(tpl) if isinstance(tpl, str) else tpl |
| if not tpl: |
| continue |
| with_t += 1 |
| for t in tpl: |
| if t["template_id"] == ss: |
| selfm += 1 |
| if bank.get(t["template_id"]) == sc: |
| ok += 1 |
| else: |
| bad += 1 |
| verdict = (bad == 0 and selfm == 0) |
| all_pass &= verdict |
| cov = 100.0 * with_t / max(n, 1) |
| print(f" {name} [{split:17s}] fg-N={n:5d} with_template={with_t:5d} ({cov:5.1f}%) " |
| f"exact_ok={ok:5d} non_exact={bad} self_matches={selfm} -> {'PASS' if verdict else 'FAIL'}") |
| return all_pass |
|
|
|
|
| if __name__ == "__main__": |
| names = sys.argv[1:] or DEFAULT |
| print("=== EXACT-MATCH SANITY CHECK ===") |
| overall = True |
| for nm in names: |
| overall &= check(nm) |
| print("\nOVERALL:", "ALL PASS" if overall else "FAIL (see above)") |
| sys.exit(0 if overall else 1) |
|
|
