Learning Depth from Past Selves: Self-Evolution Contrast for Robust Depth Estimation

TL;DR

This paper introduces SEC-Depth, a self-supervised depth estimation framework that uses self-evolution contrastive learning with latency models to improve robustness under adverse weather conditions like rain and fog.

Contribution

It proposes a novel self-evolution contrastive learning scheme utilizing intermediate training parameters to enhance depth estimation robustness in challenging weather conditions.

Findings

Significantly improves depth estimation robustness in adverse weather.

Effectively integrates with various baseline models.

Reduces manual intervention through adaptive learning objectives.

Abstract

Self-supervised depth estimation has gained significant attention in autonomous driving and robotics. However, existing methods exhibit substantial performance degradation under adverse weather conditions such as rain and fog, where reduced visibility critically impairs depth prediction. To address this issue, we propose a novel self-evolution contrastive learning framework called SEC-Depth for self-supervised robust depth estimation tasks. Our approach leverages intermediate parameters generated during training to construct temporally evolving latency models. Using these, we design a self-evolution contrastive scheme to mitigate performance loss under challenging conditions. Concretely, we first design a dynamic update strategy of latency models for the depth estimation task to capture optimization states across training stages. To effectively leverage latency models, we introduce a self-evolution contrastive Loss (SECL) that treats outputs from historical latency models as negative samples. This mechanism adaptively adjusts learning objectives while implicitly sensing weather degradation severity, reducing the needs for manual intervention. Experiments show that our method integrates seamlessly into diverse baseline models and significantly enhances robustness in zero-shot evaluations.