VMLoc: Variational Fusion For Learning-Based Multimodal Camera Localization

TL;DR

VMLoc introduces a novel variational fusion framework for multimodal camera localization, effectively combining image and depth data through a Product-of-Experts and attention mechanisms, outperforming previous methods especially with degraded inputs.

Contribution

The paper presents VMLoc, an end-to-end variational fusion approach that improves multimodal camera localization by addressing naive feature fusion and handling degraded inputs.

Findings

Outperforms previous multimodal localization methods on RGB-D datasets.

Effectively fuses image and depth modalities using a variational Product-of-Experts.

Handles degraded or missing input data robustly.

Abstract

Recent learning-based approaches have achieved impressive results in the field of single-shot camera localization. However, how best to fuse multiple modalities (e.g., image and depth) and to deal with degraded or missing input are less well studied. In particular, we note that previous approaches towards deep fusion do not perform significantly better than models employing a single modality. We conjecture that this is because of the naive approaches to feature space fusion through summation or concatenation which do not take into account the different strengths of each modality. To address this, we propose an end-to-end framework, termed VMLoc, to fuse different sensor inputs into a common latent space through a variational Product-of-Experts (PoE) followed by attention-based fusion. Unlike previous multimodal variational works directly adapting the objective function of vanilla variational auto-encoder, we show how camera localization can be accurately estimated through an unbiased objective function based on importance weighting. Our model is extensively evaluated on RGB-D datasets and the results prove the efficacy of our model. The source code is available at https://github.com/kaichen-z/VMLoc.