Domain-invariant Stereo Matching Networks

TL;DR

This paper introduces DSMNet, a stereo matching network designed to generalize across different domains by using domain normalization and a graph-based filter, outperforming existing models especially on unseen real-world data.

Contribution

The paper proposes a novel domain normalization technique and a trainable graph-based filter to enhance domain invariance in stereo matching networks, improving cross-domain generalization.

Findings

DSMNet outperforms state-of-the-art models on unseen real datasets.

It surpasses some models fine-tuned on target domains without additional training.

The approach effectively handles domain differences like color, illumination, and texture.

Abstract

State-of-the-art stereo matching networks have difficulties in generalizing to new unseen environments due to significant domain differences, such as color, illumination, contrast, and texture. In this paper, we aim at designing a domain-invariant stereo matching network (DSMNet) that generalizes well to unseen scenes. To achieve this goal, we propose i) a novel "domain normalization" approach that regularizes the distribution of learned representations to allow them to be invariant to domain differences, and ii) a trainable non-local graph-based filter for extracting robust structural and geometric representations that can further enhance domain-invariant generalizations. When trained on synthetic data and generalized to real test sets, our model performs significantly better than all state-of-the-art models. It even outperforms some deep learning models (e.g. MC-CNN) fine-tuned with test-domain data.