Distortion-Aware Monocular Depth Estimation for Omnidirectional Images

A main challenge for tasks on panorama lies in the distortion of objects among images. In this work, we propose a Distortion-Aware Monocular Omnidirectional (DAMO) dense depth estimation network to address indoor panoramas with two steps. First, introduce distortion-aware module extract calibrated semantic features from omnidirectional Specifically, exploit deformable convolution adjust its sampling grids geometric variations distorted and then utilize strip pooling sample against horizontal introduced by inverse gnomonic projection. Second, further plug-and-play spherical-aware weight matrix our objective function handle uneven distribution areas projected sphere. Experiments 360D dataset show that proposed method can effectively alleviate supervision bias caused distortion. It achieves state-of-the-art performance high efficiency.