arnaudstiegler/synthetic_us_passports_hard

Synthetic US Passports (Hard) This dataset is designed to evaluate VLMs transcription capabilities by using a well-known and straightforward document type: passports. More specifically, it requires VLMs to be robust to: tilted documents high-resolution image with a small region of interest (since the passport only takes up a part of the image) HARD VERSION ONLY: noise injected using the Augraphy package, leading to a much more difficult transcription Note: there is a "sister"… See the full description on the dataset page: https://huggingface.co/datasets/arnaudstiegler/synthetic_us_passports_ha

<iframe src="https://silytics.si/embed/item/fd99a603-235b-4ce6-9fef-49fd5fc81898" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 21 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.