2D Gaussians Spatial Transport for Point-supervised Density Regression
This is an incremental improvement for computer vision researchers working on tasks like crowd counting and landmark detection, offering a more efficient alternative to conventional optimal transport schemes.
The paper tackles the problem of point-supervised density regression in computer vision by introducing Gaussian Spatial Transport (GST), which uses Gaussian splatting to estimate pixel-annotation correspondence and compute a transport plan via Bayesian probability, eliminating iterative transport plan computation during training and improving efficiency.
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.