Hi, in your paper "Unsupervised Depth Completion with Calibrated Backprojection Layers", you trained and tested KBNet on the NYUv2 dataset. To my understanding, the original depth maps in NYUv2 are semi-dense (quasi-dense), unlike the sparse depth maps in the VOID dataset. Does this imply that KBNet can handle both sparse depth completion and semi-dense depth completion?
As I’m not very familiar with the depth completion field, I’d like to ask: When testing on NYUv2, did you directly use the semi-dense depth maps as input to the network?
Hi, in your paper "Unsupervised Depth Completion with Calibrated Backprojection Layers", you trained and tested KBNet on the NYUv2 dataset. To my understanding, the original depth maps in NYUv2 are semi-dense (quasi-dense), unlike the sparse depth maps in the VOID dataset. Does this imply that KBNet can handle both sparse depth completion and semi-dense depth completion?
As I’m not very familiar with the depth completion field, I’d like to ask: When testing on NYUv2, did you directly use the semi-dense depth maps as input to the network?