PP-OCRv4
PP-OCRv4 is the fourth generation of PaddleOCR's text detection and recognition models, released in mobile and server versions.
A release of PaddleOCR (PP-OCR). Other versions: PP-OCRv5.
- Released
- September 2023, with PaddleOCR 2.7
- Model ID
- PP-OCRv4_server_det, PP-OCRv4_mobile_det, PP-OCRv4_server_rec, PP-OCRv4_mobile_rec
- Size
- Detection: 4.70M (mobile) and 110M (server)
- License
- Apache-2.0
- Also searched as
- pp-ocrv4, ppocrv4, paddleocr 2.7
What is new in PP-OCRv4
- Over 4% more accurate than PP-OCRv3 on Chinese text and about 6% on English and digits.
- Average accuracy across four multilingual groups up by more than 8%. Latin script went from 71.57% to 80.00%.
- New detection head (PFHead), dynamic shrink ratio and the PP-LCNetV3 backbone.
- Recognition adds a Lite-Neck, GTC-NRTR guided training, multi-scale training and DKD distillation.
- First release with both mobile and server versions of the models.
PP-OCRv4 benchmarks
| Benchmark | Score | Note |
|---|---|---|
| English recognition accuracy | 70.1% | PP-OCRv3: 64.04%. PaddleOCR's own set |
| Latin-script recognition accuracy | 80.00% | PP-OCRv3: 71.57% |
| End-to-end Chinese Hmean | 62.24% | PP-OCRv3: 57.99% |
Scores as published by the authors. We have not rerun them.
Use PP-OCRv4 through an API
OCRRouter runs a different release of PaddleOCR (PP-OCR) as paddle/paddleocr.
1PaddleOCR-VLOpen-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
paddle/paddleocr-vl6 credits / page2GLM-OCRSmall open-source OCR model from Z.ai with layout regions. MIT licensed.zai/glm-ocr6 credits / pagePaddleOCR (PP-OCR) versions
- PaddleOCR (PP-OCR): all versions, price, speed and features on OCRRouter
- PP-OCRv5 (May 20, 2025, with PaddleOCR 3.0)
- PP-OCRv4 (September 2023, with PaddleOCR 2.7), this page
Questions
Sources
Last updated .