Hansda/captcha-dataset

Synthetic CAPTCHA Dataset Synthetic distorted-text CAPTCHA images with exact character-string labels, used to fine-tune Hansda/minicpm-v4_6-captcha-lora. Contents Format: HuggingFace imagefolder (image, text columns), with metadata.jsonl per split. Splits: train (10,000) and test (2,000) — 12,000 images total. Labels: 4–7 characters from ABCDEFGHJKLMNPQRSTUVWXYZabcdefghjkmnpqrstuvwxyz23456789 (visually ambiguous I O 0 1 l are excluded). Generation… See the full description on the dataset page: https://huggingface.co/datasets/Hansda/captcha-dataset.