Community Reproductions, Integrations, and Failure Cases

[Yuliang-Liu]

We are collecting reproducible downstream integrations and independent evaluations of MonkeyOCRv2.

Please share:

• the task and dataset
• the downstream architecture
• training and evaluation settings
• checkpoints or reproducible code
• positive results and failure cases

We are especially interested in text recognition, formula recognition, detection, segmentation, multilingual documents, historical documents, and document understanding.

Current results show consistent gains across seven document tasks, but we hope the community can reproduce, challenge, and improve these findings.

Paper:
https://huggingface.co/papers/2607.11562

Code:
https://github.com/Yuliang-Liu/MonkeyOCRv2