We introduce GVIT, a classification framework that abandons conventional pixel or patch grid input representations in favor of a compact set of learnable 2D Gaussians. Each image is encoded as a few hundred Gaussians whose positions, scales, orientations, colors, and opacities are optimized jointly with a ViT classifier trained on top of these representations. We reuse the classifier gradients as constructive guidance, steering the Gaussians toward class-salient regions while a differentiable renderer optimizes an image reconstruction loss. We demonstrate that by 2D Gaussian input representations coupled with our GVIT guidance, using a relatively standard ViT architecture, closely matches the performance of a traditional patch-based ViT, reaching a 76.9% top-1 accuracy on Imagenet-1k using a ViT-B architecture.
@article{arxiv.2506.23532,
title = {GViT: Representing Images as Gaussians for Visual Recognition},
author = {Jefferson Hernandez and Ruozhen He and Guha Balakrishnan and Alexander C. Berg and Vicente Ordonez},
journal= {arXiv preprint arXiv:2506.23532},
year = {2025}
}