| """Package the NEW (v2.0) images that survived QC into Evren-ready ZIP parts (<=300 MB, every part mixes classes). |
| Writes meta2/manifest_v2_new.csv and out2/*.zip.""" |
| import csv, glob, os, zipfile |
| from collections import Counter, defaultdict |
|
|
| D2 = '/home/user/hemato/dataset2'; OUT = '/home/user/hemato/out2'; META = '/home/user/hemato/meta2' |
| LIMIT = 300_000_000 |
| os.makedirs(OUT, exist_ok=True); os.makedirs(META, exist_ok=True) |
| removed = {r['file'] for r in csv.DictReader(open('/home/user/hemato/qc2_removed.csv'))} |
| R = [r for f in sorted(glob.glob('/home/user/hemato/manifest2/*.csv')) for r in csv.DictReader(open(f)) |
| if r['split'] in ('train', 'val', 'test') and r['file'] not in removed and os.path.exists(os.path.join(D2, r['file']))] |
| with open(f'{META}/manifest_v2_new.csv', 'w', newline='', encoding='utf-8-sig') as f: |
| w = csv.DictWriter(f, fieldnames=list(R[0])); w.writeheader(); w.writerows(R) |
| NAMES = {'cls': 'HEMATO-CLS', 'all': 'HEMATO-ALL', 'malaria': 'HEMATO-MALARIA'} |
| for cfg, name in NAMES.items(): |
| files = sorted(r['file'] for r in R if r['config'] == cfg) |
| if not files: continue |
| by = defaultdict(list) |
| for f in files: by[f.split('/')[2]].append(f) |
| allf = [f for lab in sorted(by) for f in sorted(by[lab])] |
| total = sum(os.path.getsize(os.path.join(D2, f)) for f in allf) |
| n = max(1, -(-total // LIMIT)) |
| for i in range(n): |
| part = allf[i::n] |
| zn = f'{OUT}/{name}-v2_part{i + 1:02d}-of-{n:02d}.zip' |
| with zipfile.ZipFile(zn, 'w', zipfile.ZIP_STORED) as z: |
| for f in part: z.write(os.path.join(D2, f), f.split('/', 1)[1]) |
| labs = Counter(f.split('/')[2] for f in part) |
| print(os.path.basename(zn), round(os.path.getsize(zn) / 1e6), 'MB', len(part), 'files', len(labs), 'classes') |
| det = [r for r in R if r['config'] == 'det'] |
| if det: |
| zn = f'{OUT}/HEMATO-DET-v2.zip' |
| with zipfile.ZipFile(zn, 'w', zipfile.ZIP_DEFLATED) as z: |
| for r in det: |
| img = os.path.join(D2, r['file']); lbl = img.replace('/images/', '/labels/').rsplit('.', 1)[0] + '.txt' |
| z.write(img, os.path.relpath(img, f'{D2}/det')); z.write(lbl, os.path.relpath(lbl, f'{D2}/det')) |
| z.writestr('data.yaml', 'path: .\ntrain: images/train\nval: images/val\ntest: images/test\nnc: 3\nnames: [Platelets, RBC, WBC]\n') |
| print(os.path.basename(zn), round(os.path.getsize(zn) / 1e6), 'MB', len(det), 'images') |
| print(Counter((r['config'], r['split']) for r in R)) |
|
|
