24 upvotes
OvisOCR3 packs OCR, KIE, and table stitching into 0.8B model
What is OvisOCR3 Technical Report?
OvisOCR3 is a 0.8-billion-parameter model that can read printed or handwritten text, pull out key data, caption images, and rebuild tables that span multiple pages, all with a single network. It was trained on a mix of real documents and synthetic data to cover a wide variety of document types.
Who it is for
Anyone building document-automation pipelines or low-code AI apps can replace a stack of specialized OCR tools with this one model.
How to try it
Visit the Hugging Face repo at https://huggingface.co/ovisocr3