Monocular Depth Estimation from a Single Infrared Image

Thermal infrared imaging is attracting much attention due to its strength against illuminance variation. However, because of the spectral difference between thermal images and RGB images, existing research on self-supervised monocular depth estimation has performance limitations. Therefore, in this study, we propose a novel Self-Guided Framework using Pseudolabel predicted from images. Our proposed framework, which solves problem appearance matching loss transfers high accuracy network by comparing low- high-level pixels. Furthermore, Patch-NetVLAD Loss, strengthens local detail global context information map locally patch-level descriptors. Finally, introduce an Image Matching Loss estimate more accurate enhancing Pseudolabel. We demonstrate that framework shows significant improvement even when applied various networks KAIST Multispectral Dataset.