# GN-Net: The Gauss-Newton Loss for Multi-Weather Relocalization

## TL;DR

GN-Net introduces a Gauss-Newton loss to train weather-invariant features for direct image alignment, significantly improving relocalization robustness under challenging weather and lighting conditions.

## Contribution

The paper proposes a novel Gauss-Newton loss for training deep features that are invariant to weather changes, enhancing direct relocalization performance.

## Key findings

- Outperforms state-of-the-art methods in robustness against weather variations.
- Demonstrates effectiveness on both simulated and real-world datasets.
- Provides an open benchmark for relocalization under different weather conditions.

## Abstract

Direct SLAM methods have shown exceptional performance on odometry tasks. However, they are susceptible to dynamic lighting and weather changes while also suffering from a bad initialization on large baselines. To overcome this, we propose GN-Net: a network optimized with the novel Gauss-Newton loss for training weather invariant deep features, tailored for direct image alignment. Our network can be trained with pixel correspondences between images taken from different sequences. Experiments on both simulated and real-world datasets demonstrate that our approach is more robust against bad initialization, variations in day-time, and weather changes thereby outperforming state-of-the-art direct and indirect methods. Furthermore, we release an evaluation benchmark for relocalization tracking against different types of weather. Our benchmark is available at https://vision.in.tum.de/gn-net.