PaddleOCR-VL-1.5
PaddleOCR-VL-1.5 is an earlier available 0.9B document model. It improves recognition under skew, warped pages and uneven lighting, and adds support for Tibetan and Bengali.
Model design and access
The release adds polygon-based text detection and document recognition improvements. Its broader pipeline also provides header handling and cross-page table merging. These pipeline operations should not be assumed to come from a direct call to the vision-language checkpoint alone.PaddleOCR-VL-1.5 official model card
Practical use
Practical guidance. Retain this version when an existing PaddleOCR integration has been reviewed against it. Use photographs as well as clean scans in a migration comparison. Keep preprocessing settings fixed so changes in deskewing or cropping do not hide the effect of the model update.PaddleOCR-VL-1.5 official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Apache 2.0PaddleOCR-VL-1.5 official model card |
| Parameters | 0.9 billionPaddleOCR-VL-1.5 official model card |
| Tasks | Document parsing, seal recognition and text spottingPaddleOCR-VL-1.5 official model card |
Related models
- PaddleOCR-VL-1.6
Baidu ยท PaddleOCR-VL
A 0.9-billion-parameter document vision-language model with improved training for difficult document regions.
Launch date Downloadable