Depth Map Decomposition for Monocular Depth Estimation

AbstractWe propose a novel algorithm for monocular depth estimation that decomposes metric map into normalized and scale features. The proposed network is composed of shared encoder three decoders, called G-Net, N-Net, M-Net, which estimate gradient maps, map, respectively. M-Net learns to depths more accurately using relative features extracted by G-Net N-Net. has the advantage it can use datasets without labels improve performance estimation. Experimental results on various demonstrate not only provides competitive state-of-the-art algorithms but also yields acceptable even when small amount data available its training.KeywordsMonocular estimationRelative estimationDepth decomposition