Intrinsic Phase-Preserving Networks for Depth Super Resolution

• DOI: 10.1609/aaai.v38i2.27883
• Publication Date: 2024-03-25T09:07:59Z
• Authors: Xuanhong Chen; Hang Wang; Jialiang Chen; Kairui Feng; Jinfan Liu; Xiaohang Wang; Weimin Zhang; Bingbing Ni

Depth map super-resolution (DSR) plays an indispensable role in 3D vision. We discover non-trivial spectral phenomenon: the components of high-resolution (HR) and low-resolution (LR) depth maps manifest same intrinsic phase, phase RGB is a superset them, which suggests that phase-aware filter can assist precise use cues. Motivated by this, we propose phase-preserving DSR paradigm, named IPPNet, to fully exploit inter-modality collaboration mutually guided way. In nutshell, novel Phase-Preserving Filtering Module (PPFM) developed generate dynamic filters according LR flow out erroneous noisy contained then conduct enhancement via modulation phase-preserved signal. By stacking multiple PPFM blocks, proposed IPPNet capable reaching highly competitive restoration performance. Extensive experiments on various benchmark datasets, e.g., NYU v2, RGB-D-D, reach SOTA performance also well demonstrate validity scheme. Code: https://github.com/neuralchen/IPPNet/.