PP-OCRv6 released: 1.5M-34.5M params, outperforms billion-scale VLMs

Arxiv CS.CV (computer vision)

Baidu's PP-OCRv6 model series scales from 1.5M to 34.5M parameters, achieving +4.9% detection and +5.1% recognition accuracy over prior PP-OCR. It surpasses billion-scale VLMs on OCR tasks while being lightweight enough for browser and edge deployment.

PaddleOCROCR

See 2 more sources

🚀PP-OCRv6 is officially released !1 day agoKokaOPDiscuss

SenseNova U1 dropped an infographic-specific finetune3 days agoMatakotightDiscuss

Arxiv CS.CV (computer vision)·Yubo ZhangXueqing WangManhui LinYue ZhangPenglongyi DengTing SunTingquan GaoZelun ZhangJiaxuan LiuChangda ZhouHongen LiuSuyin LiangCheng CuiYi LiuDianhai YuYanjun Ma

2 days ago