PaddleOCR-VL-1.6: under-optimized region refinement boosts document parsing

Arxiv CS.CV (computer vision)

PaddleOCR-VL-1.6 is an upgraded 0.9B-parameter document parsing model built on PaddleOCR-VL-1.5, focusing on under-optimized region refinement and progressive post-training to reduce remaining errors. The model is available on HuggingFace from PaddlePaddle.

PaddleOCROCR

See 2 more sources

PaddlePaddle/PaddleOCR-VL-1.629 days agoPaddlePaddle

RT @LeRobotHF: As many of you were interested in the technical details of the model, here is a...18 days agoYann LeCun

Arxiv CS.CV (computer vision)·Zelun ZhangHongen LiuSuyin LiangYubo ZhangYiqing XiangJiaxuan LiuTing SunManhui LinYue ZhangChangda ZhouTingquan GaoCheng CuiYi LiuDianhai YuYanjun Ma

Jun 3, 4:00 AM