Tunable Soft Equivariance with Guarantees
2026-03-27 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionMachine Learning
AI summaryⓘ
The authors address the problem that computer vision models often don't perfectly handle changes like rotation or translation, which are called equivariance. They introduce a new way to gently control how much the model is equivariant by adjusting the model's settings into a special space. This method works with already-trained models and comes with guarantees about its effects. They tested it on popular models like ViT and ResNet in tasks such as image classification and found it improved performance while making the models better at handling transformations.
equivariancecomputer visionmodel weightspre-trained architectureViTResNetimage classificationsemantic segmentationImageNet benchmarktrajectory prediction
Authors
Md Ashiqur Rahman, Lim Jun Hao, Jeremiah Jiang, Teck-Yian Lim, Raymond A. Yeh
Abstract
Equivariance is a fundamental property in computer vision models, yet strict equivariance is rarely satisfied in real-world data, which can limit a model's performance. Controlling the degree of equivariance is therefore desirable. We propose a general framework for constructing soft equivariant models by projecting the model weights into a designed subspace. The method applies to any pre-trained architecture and provides theoretical bounds on the induced equivariance error. Empirically, we demonstrate the effectiveness of our method on multiple pre-trained backbones, including ViT and ResNet, across image classification, semantic segmentation, and human-trajectory prediction tasks. Notably, our approach improves the performance while simultaneously reducing equivariance error on the competitive ImageNet benchmark.