PolyOCR-Venus: Unified OCR Foundation Models for Text-Centric Visual Intelligence
PolyOCR introduces multi-scale OCR foundation models and a corrected OCRBench, reporting competitive document and scene-text results.
PolyOCR is a family of unified OCR foundation models designed to recognize, localize, parse, and reason over text in complex images. The report describes a shared instruction-following framework, a large-scale data engine, and Competence-Guided Policy Optimization combining verifier-based group relative policy optimization with routed on-policy distillation. The authors also release OCRBench v2.1 with manually verified annotation corrections and task-aligned metrics. Across OCRBench v2.1, CC-OCR, an in-house key-information extraction benchmark, OmniDocBench v1.6, and MDPBench, PolyOCR is reported as state of the art or highly competitive.
- PolyOCR is a multi-scale family of unified OCR foundation models.
- A data engine turns heterogeneous visuals into verified OCR supervision.
- Competence-Guided Policy Optimization mixes GRPO with on-policy distillation.
- OCRBench v2.1 corrects annotations and aligns scoring metrics.
- Results are competitive on OCRBench, CC-OCR, OmniDocBench, and MDPBench.
From the original article
1 min read at huggingface.co
Optical Character Recognition (OCR) is evolving from plain-text transcription toward general visual intelligence, requiring models to recognize, localize, and reason over textual information in complex visual environments. However, existing OCR systems often excel at only some tasks and struggle to balance recognition, parsing, and reasoning across scenarios.
By GuangJian Team, Kaili Huang, Yongshuo Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.