FoVA-Depth: Field-of-View Agnostic Depth Estimation for Cross-Dataset Generalization

• DOI: 10.48550/arxiv.2401.13786
• Publication Date: 2024-01-01
• Authors: Daniel Lichy; Hang Su; Abhishek Badki; Jan Kautz; Orazio Gallo
• Keywords: Ground truth; Field of view; Intuition; Pinhole camera; Depth of field; Depth map

Wide field-of-view (FoV) cameras efficiently capture large portions of the scene, which makes them attractive in multiple domains, such as automotive and robotics. For applications, estimating depth from images is a critical task, therefore, amount ground truth (GT) data available. Unfortunately, most GT for pinhole cameras, making it impossible to properly train estimation models large-FoV cameras. We propose first method stereo model on widely available data, generalize captured with larger FoVs. Our intuition simple: warp training canonical, representation augment allow single network reason about diverse types distortions that otherwise would prevent generalization. show strong generalization ability our approach both indoor outdoor datasets, was not possible previous methods.