Abstract
We present TemporalStereo, a coarse-to-fine based online stereo matching network which is highly efficient, and able to effectively exploit the past geometry and context information to boost the matching accuracy. Our network leverages sparse cost volume and proves to be effective when a single stereo pair is given, however, its peculiar ability to use spatio-temporal information across frames allows TemporalStereo to alleviate problems such as occlusions and reflective regions while enjoying high efficiency also in the case of stereo sequences. Notably our model trained, once with stereo videos, can run in both single-pair and temporal ways seamlessly. Experiments show that our network relying on camera motion is even robust to dynamic objects when running on videos. We validate TemporalStereo through extensive experiments on synthetic (SceneFlow, TartanAir) and real (KITTI 2012, KITTI 2015) datasets. Detailed results show that our model achieves state-of-the-art performance on any of these datasets. Code is available at \url{this https URL}.
Abstract (translated)
URL
https://arxiv.org/abs/2211.13755
