Differentiable Particle Filters through Conditional Normalizing Flow

TL;DR

This paper introduces a novel approach using conditional normalizing flows to enhance proposal and dynamic models in differentiable particle filters, improving their flexibility and performance in visual tracking tasks.

Contribution

It proposes integrating conditional normalizing flows into differentiable particle filters to create more expressive proposal and dynamic models, addressing limitations of existing methods.

Findings

Improved tracking accuracy in visual tasks

Enhanced flexibility of proposal distributions

More expressive dynamic models

Abstract

Differentiable particle filters provide a flexible mechanism to adaptively train dynamic and measurement models by learning from observed data. However, most existing differentiable particle filters are within the bootstrap particle filtering framework and fail to incorporate the information from latest observations to construct better proposals. In this paper, we utilize conditional normalizing flows to construct proposal distributions for differentiable particle filters, enriching the distribution families that the proposal distributions can represent. In addition, normalizing flows are incorporated in the construction of the dynamic model, resulting in a more expressive dynamic model. We demonstrate the performance of the proposed conditional normalizing flow-based differentiable particle filters in a visual tracking task.