Deformable spatial propagation network for depth completion
This addresses depth completion for autonomous driving applications, representing an incremental improvement over existing convolutional spatial propagation networks.
The paper tackles the problem of depth completion from sparse measurements by proposing a deformable spatial propagation network (DSPN) that adaptively generates different receptive fields and affinity matrices for each pixel, achieving state-of-the-art performance on the KITTI benchmark.
Depth completion has attracted extensive attention recently due to the development of autonomous driving, which aims to recover dense depth map from sparse depth measurements. Convolutional spatial propagation network (CSPN) is one of the state-of-the-art methods in this task, which adopt a linear propagation model to refine coarse depth maps with local context. However, the propagation of each pixel occurs in a fixed receptive field. This may not be the optimal for refinement since different pixel needs different local context. To tackle this issue, in this paper, we propose a deformable spatial propagation network (DSPN) to adaptively generates different receptive field and affinity matrix for each pixel. It allows the network obtain information with much fewer but more relevant pixels for propagation. Experimental results on KITTI depth completion benchmark demonstrate that our proposed method achieves the state-of-the-art performance.