基于稀疏自注意力的偏振表面法线估计
Shape from polarization based on sparse self-attention
于智超 1万振华 2赵开春1
作者信息
- 1. 清华大学 精密仪器系,北京 100084
- 2. 广西大学 机械工程学院,广西 南宁 530004
- 折叠
摘要
对于物体表面法线的准确估计在各种计算机视觉任务中扮演着重要角色.采用基于物理的偏振表面法线估计方法存在诸多限制,限制了其实用性.相反,基于深度学习的偏振表面法线估计方法在精度和适用性上均超越了基于物理的方法,为了进一步提高基于深度学习的偏振表面法线估计精度,以使其能适用于更多的实际任务,本文提出了一种新的方法.首先,结合斯托克斯矢量提出了一种新的偏振信息表示方法,旨在提高模型对偏振物理先验信息的提取能力.随后结合基于双层路由的稀疏自注意力机制,以改进模型对全局上下文信息的感知能力,来更好地对局部偏振信息消歧.在DeepSfP数据集和自建测试数据进行测试.实验结果表明:提出的方法在DeepSfP数据集上平均角度误差为13.37°,并在精度和角度误差等所有测试指标上均优于现有方法,证明了该方法在估计法线效果方面的显著改进.通过引入新的偏振信息表示方法和稀疏自注意力机制,我们的方法提高了偏振表面法线估计的精度和适用性,为实际任务的应用提供了更强的支持.
Abstract
Accurate estimation of surface normal plays a vital role in various computer vision tasks.Physi-cally-based shape from polarization methods have limitations restricting their applications.Conversely,learning-based shape from polarization methods outperform physical methods in both accuracy and applica-bility.To further improve the accuracy of shape from polarization and make it applicable to a broader range of practical tasks,we proposed a novel method.First,we introduced a new polarization information repre-sentation combining Stokes vectors,enhancing the model's ability to extract polarization physical prior in-formation.Then,we integrated a bi-level routing sparse self-attention mechanism to improve the model's perception of global contextual information,enabling better disambiguation of local polarization informa-tion.Testing on the DeepSfP dataset and out test data,experimental results demonstrate our proposed method achieves an average angular error of 13.37° on the DeepSfP dataset,outperforming existing meth-ods in all tested metrics including accuracy and angular error.This indicates a significant improvement in normal estimation effectiveness with our proposed method.By introducing a novel polarization information representation and sparse self-attention mechanism,our approach enhances the accuracy and applicability of polar surface normal estimation,providing stronger support for practical task applications.
关键词
偏振信息表达/稀疏自注意力机制/偏振表面法线估计Key words
polarization information representation/sparse self-attention mechanism/shape from polar-ization引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据