Datasets:
Commit History
25 days ago
-
Add pilot dataset: 4,000 shaped Devanagari samples + masks, fonts, generator (part 14)
Verified signature4,000 synthetic 512x512 image/text pairs with HarfBuzz-shaped Devanagari, exact transcriptions and glyph masks. - 72 train words, 20 held out and never rendered - 22 OFL fonts, all audited for coverage and conjunct formation - font em size >=32px enforced (measured SD1.5 VAE floor) - generator included; dataset is reproducible from seed 4224e04bb