ACDNet: Adaptively Combined Dilated Convolution for Monocular Panorama Depth Estimation

TL;DR

ACDNet introduces an adaptively combined dilated convolution approach with channel-wise attention for improved monocular panoramic depth estimation, effectively handling distortion and expanding receptive fields.

Contribution

The paper proposes a novel ACDNet that adaptively combines dilated convolutions with channel-wise attention to enhance depth estimation accuracy for panoramic images.

Findings

Outperforms current SOTA methods on three datasets

Effectively captures cross-channel contextual information

Handles distortion in equirectangular projection

Abstract

Depth estimation is a crucial step for 3D reconstruction with panorama images in recent years. Panorama images maintain the complete spatial information but introduce distortion with equirectangular projection. In this paper, we propose an ACDNet based on the adaptively combined dilated convolution to predict the dense depth map for a monocular panoramic image. Specifically, we combine the convolution kernels with different dilations to extend the receptive field in the equirectangular projection. Meanwhile, we introduce an adaptive channel-wise fusion module to summarize the feature maps and get diverse attention areas in the receptive field along the channels. Due to the utilization of channel-wise attention in constructing the adaptive channel-wise fusion module, the network can capture and leverage the cross-channel contextual information efficiently. Finally, we conduct depth estimation experiments on three datasets (both virtual and real-world) and the experimental results demonstrate that our proposed ACDNet substantially outperforms the current state-of-the-art (SOTA) methods. Our codes and model parameters are accessed in https://github.com/zcq15/ACDNet.