Digging Into Self-Supervised Monocular Depth Estimation

TL;DR

This paper introduces simple yet effective improvements to self-supervised monocular depth estimation, achieving state-of-the-art results by focusing on robust loss functions and sampling methods rather than complex architectures.

Contribution

The paper presents a set of novel, simple design choices that significantly enhance self-supervised monocular depth estimation performance.

Findings

Achieved state-of-the-art results on KITTI benchmark.

Proposed a minimum reprojection loss for occlusion handling.

Introduced a full-resolution multi-scale sampling method.

Abstract

Per-pixel ground-truth depth data is challenging to acquire at scale. To overcome this limitation, self-supervised learning has emerged as a promising alternative for training models to perform monocular depth estimation. In this paper, we propose a set of improvements, which together result in both quantitatively and qualitatively improved depth maps compared to competing self-supervised methods. Research on self-supervised monocular training usually explores increasingly complex architectures, loss functions, and image formation models, all of which have recently helped to close the gap with fully-supervised methods. We show that a surprisingly simple model, and associated design choices, lead to superior predictions. In particular, we propose (i) a minimum reprojection loss, designed to robustly handle occlusions, (ii) a full-resolution multi-scale sampling method that reduces visual artifacts, and (iii) an auto-masking loss to ignore training pixels that violate camera motion assumptions. We demonstrate the effectiveness of each component in isolation, and show high quality, state-of-the-art results on the KITTI benchmark.