Towards Generalized Multimodal Homography Estimation

TL;DR

This paper introduces a novel training data synthesis method and a specialized network to enhance the robustness and generalization of homography estimation across different modalities and domains.

Contribution

It presents a new data synthesis approach generating diverse, structurally preserved image pairs from a single image, and a network that leverages cross-scale info and decouples color for better accuracy.

Findings

Synthetic data improves cross-domain generalization.

The proposed network outperforms existing methods.

Enhanced robustness across unseen modalities.

Abstract

Supervised and unsupervised homography estimation methods depend on image pairs tailored to specific modalities to achieve high accuracy. However, their performance deteriorates substantially when applied to unseen modalities. To address this issue, we propose a training data synthesis method that generates unaligned image pairs with ground-truth offsets from a single input image. Our approach renders the image pairs with diverse textures and colors while preserving their structural information. These synthetic data empower the trained model to achieve greater robustness and improved generalization across various domains. Additionally, we design a network to fully leverage cross-scale information and decouple color information from feature representations, thus improving estimation accuracy. Extensive experiments show that our training data synthesis method improves generalization performance. The results also confirm the effectiveness of the proposed network.