2D Gaussians Spatial Transport for Point-supervised Density Regression

TL;DR

This paper presents Gaussian Spatial Transport (GST), a new framework using Gaussian splatting for efficient point-supervised density regression in computer vision, improving over traditional methods by eliminating iterative transport computations.

Contribution

The paper introduces GST, a novel Gaussian splatting-based transport method that simplifies and accelerates point-supervised density regression tasks in vision applications.

Findings

GST outperforms conventional schemes in efficiency

Effective in crowd counting and landmark detection

Eliminates iterative transport plan computation

Abstract

This paper introduces Gaussian Spatial Transport (GST), a novel framework that leverages Gaussian splatting to facilitate transport from the probability measure in the image coordinate space to the annotation map. We propose a Gaussian splatting-based method to estimate pixel-annotation correspondence, which is then used to compute a transport plan derived from Bayesian probability. To integrate the resulting transport plan into standard network optimization in typical computer vision tasks, we derive a loss function that measures discrepancy after transport. Extensive experiments on representative computer vision tasks, including crowd counting and landmark detection, validate the effectiveness of our approach. Compared to conventional optimal transport schemes, GST eliminates iterative transport plan computation during training, significantly improving efficiency. Code is available at https://github.com/infinite0522/GST.