Description
PaddleOCR-VL-1.6 is an upgraded document parsing model developed to improve upon its predecessor, PaddleOCR-VL-1.5. This model introduces a region-aware data optimization framework that identifies and enhances weak regions from the previous version, thereby improving the reliability of supervision signals. Additionally, it employs a progressive post-training recipe that utilizes curated data selection and reinforcement learning to optimize model performance.
Read More
PaddleOCR-VL-1.6 has achieved a new state-of-the-art score of 96.33% on the OmniDocBench v1.6, setting new records on OmniDocBench v1.5 and Real5-OmniDocBench. It demonstrates strong competitiveness against top-tier vision-language models (VLMs). The model architecture is fully compatible with PaddleOCR-VL-1.5, allowing for zero-cost plug-and-play migration.
The model offers significant improvements in various recognition tasks, including table recognition, Chinese ancient document recognition, and Chinese rare character recognition. It also enhances seal/stamp recognition, text spotting, and chart recognition across diverse scenarios.
PaddleOCR-VL-1.6 supports seamless inference via the transformers library, facilitating comprehensive text spotting and recognition of complex elements such as formulas, tables, charts, and seals. Users are encouraged to utilize the official method for inference, which is faster and supports page-level document parsing.
- CategoryAI Models & LLMs
- Pricing ModelFree
- DeploymentCloud
- PlatformsWeb
PaddleOCR-VL-1.6 Highlights
Region-aware data optimization framework
Progressive post-training recipe
State-of-the-art accuracy on OmniDocBench
Compatibility with PaddleOCR-VL-1.5
Enhanced table and text recognition
Chinese ancient document recognition
Seal and stamp recognition
Supports transformers library for inference
Getting Started with PaddleOCR-VL-1.6
Access model: Visit Hugging Face
Authenticate: Ensure proper credentials
Set up environment: Install PaddlePaddle framework
Integrate via API: Use PaddleOCR CLI or Python API
Optimise: Use flash-attn to boost performance
PaddleOCR-VL-1.6's Use Cases
- Document Parsing
- Table Recognition
- Text Spotting
- Seal Recognition
- Chinese Document Parsing
