Context-Enhanced Stereo Transformer

Stereo depth estimation is of great interest for computer vision research. However, existing methods struggles to generalize and predict reliably in hazardous regions, such as large uniform regions. To overcome these limitations, we propose Context Enhanced Path (CEP). CEP improves the generalization robustness against common failure cases solutions by capturing long-range global information. We construct our stereo model, Transformer (CSTR), plugging into state-of-the-art method Transformer. CSTR examined on distinct public datasets, Scene Flow, Middlebury-2014, KITTI-2015, MPI-Sintel. find outperforms prior approaches a margin. For example, zero-shot synthetic-to-real setting, best competing Middlebury-2014 dataset 11 $$\%$$ . Our extensive experiments demonstrate that information critical matching task successfully captures information( $$^1$$ Code available at: github.com/guoweiyu/Context-Enhanced-Stereo-Transformer ).