AI Paper Atlas

Research map / Detection and segmentation / CNN features

Tunable Soft Equivariance with Guarantees

Md Ashiqur Rahman, Lim Jun Hao, Jeremiah Jiang, Teck-Yian Lim, Raymond A. Yeh

CVPR 2026 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add soft equivariance projection to ViTViT · ImageNet-1Ktop-1 accuracy: 81.67 -> 82.28Improved
add soft equivariance projection to DINOv2DINOv2 · ImageNet-1Ktop-1 accuracy: 84.27 -> 85.31Improved
add soft equivariance projection to ResNet-50ResNet-50 · ImageNet-1Kinvariance error (x10^-2): 00.24 -> 00.11Improved
add soft equivariance projection to ResNet-50ResNet-50 · CIFAR-10invariance error (x10^-2): 27.29 -> 24.20Improved
use canonicalizer instead of soft equivarianceViT · ImageNet-1Ktop-1 accuracy: 81.67 -> 76.51Hurt
use canonicalizer instead of soft equivarianceResNet-50 · CIFAR-100top-1 accuracy: 83.10 -> 72.90Hurt
add soft equivariance projection to DINOv2 for segmentationDINOv2 · PASCAL VOCmIoU: 89.57 -> 89.48No clear effect
add soft equivariance projection to ViT for segmentationViT · PASCAL VOCequivariance error (x10^-2): 12.31 -> 11.12Improved
use smooth cut-off instead of hard threshold in projection (ablation)ViT · PASCAL VOCmIoU: 73.92 -> 74.78Improved
use smooth cut-off instead of hard threshold in projection (ablation)ViT · PASCAL VOCequivariance error (x10^-2): 11.74 -> 11.12Improved
use Schur decomposition instead of SVD to build projection (ablation)projection operator · 14x14 inputruntime (s): 8.9 · 10^2 -> 2.5 · 10^-1Improved
add soft equivariance projection to autoregressive transformerAutoregressive transformer · ZARA1 trajectoryequivariance error (x10^-2): 1.14 -> 0.39Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy