arxivcs.CV2026-07-13
MonkeyOCRv2: A Visual-Text Foundation Model for Document AI
Yuliang Liu, Zhang Li, Ziyang Zhang, Shuo Zhang, Qiang Liu, Jiajun Song, et al.
Mainstream visual encoders are pretrained on natural images and cannot be effectively applied to document images without document-oriented adaptation, as dense text and fine-grained character strokes demand character-level visual perception. We present MonkeyOCRv2, a visual-text…