Recapture: Create realistic scans and photos from digital documents to diversify AI training sets
Blog post from Hugging Face
Recapture is an open-source Nutrient tool for expanding document-model training data by turning labeled digital pages into realistic scans and phone photos, addressing the shortage of varied real-world examples such as receipts, IDs, and scanned forms. It preserves layout annotations across generated versions, reducing the need to label each image or print and photograph documents manually. Its fast, GPU-free scan mode adds printer and scanner artifacts, while photo mode simulates paper folds and curls, desk surfaces, lighting, shadows, and virtual-camera effects; settings can be fixed or randomized to match a desired dataset mix. Users provide page images and layout labels, configure and preview the transformations, then render and export the results in formats such as COCO, with variants of each page kept in the same data split. Recapture also supports document-parser output and coding-agent workflows, enabling users to generate datasets through natural-language instructions.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.