Bilateral Grid Learning for Stereo Matching Networks

• DOI: 10.48550/arxiv.2101.01601
• Publication Date: 2021-01-01
• Authors: Bin Xu; Yuhua Xu; Xiaoli Yang; Wei Jia; Yulan Guo
• Keywords: Upsampling; Code (set theory)

Real-time performance of stereo matching networks is important for many applications, such as automatic driving, robot navigation and augmented reality (AR). Although significant progress has been made in recent years, it still challenging to balance real-time accuracy. In this paper, we present a novel edge-preserving cost volume upsampling module based on the slicing operation learned bilateral grid. The layer parameter-free, which allows us obtain high quality resolution from low-resolution under guide guidance map efficiently. proposed can be seamlessly embedded into existing networks, GCNet, PSMNet, GANet. resulting are accelerated several times while maintaining comparable Furthermore, design network (named BGNet) module, outperforms published deep well some complex KITTI datasets. code available at https://github.com/YuhuaXu/BGNet.