GFI-Net: Global Feature Interaction Network for Monocular Depth Estimation

Abstract

Monocular depth estimation is one of the most important tasks in the field of intelligent perception. However, there are still several problems, such as insufficient estimation accuracy, inaccurate localization of details and depth discontinuity in broad planes at long distances. To solve these problems, we propose the Global Feature Interaction Network (GFI-Net), which aims to utilize geometric features, such as object locations and vanishing points, on a global scale. The global attention mechanism module in the network improves the performance of the depth estimation by mitigating information reduction and amplifying global interaction representations. Furthermore, transformer is also introduced to the encoder module to mitigate information loss and obtain higher estimation accuracy. Experiment results on the NYU-Depth-v2 dataset and the KITTI dataset show that our model achieves the state-of-the-art performance with complete detail recoveries and depth continuation in broad planes at long distances.