AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Stabilizing Feature Geometry in Noisy Pretrained Models for Robust Downstream Tasks

Quanyu Zhang, Zhongyi Han, Hao Sun, Yongshun Gong, Xiaoyan Wang, Yilong Yin, Shuo Li

CVPR 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias

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 Feature Geometry Stabilization on top of MLP tuningMLP probe on noisy-pretrained backbones · DomainNet OOD, ViT-L Laion-2Baccuracy: 0.7359 -> 0.7662Improved
add FGS instead of NMTune spectral regularizationResNetv2-152x2 ImageNet-21K · DomainNet OODaccuracy: 0.5052 -> 0.5408Improved
NMTune spectral amplification versus plain MLPResNetv2-152x2 ImageNet-21K · DomainNet OODaccuracy: 0.5126 (MLP) -> 0.5052 (NMTune)Hurt
add FGS on in-domain tasksResNetv2-152x2 ImageNet-21K · 8 in-domain datasetsaccuracy: 0.8523 -> 0.8665 (MLP), 0.8511 (NMTune)Improved
add Feature Perturbation Consistency alone (ablation)MLP probe · OOD DomainNet Sketch, ResNet-50 on noisy ImageNet-1K and YFCC15Maccuracy: FPC beats VAR by 1.29% and FCD by 0.88% at 30% noise on ImageNet-1KImproved
add VAR alone or FCD alone (ablation)MLP probe · OOD DomainNet Sketch, ResNet-50accuracyImproved
combine FPC, VAR and FCD (ablation)FGS · OOD DomainNet Sketch, ResNet-50 all noise levelsaccuracyImproved
use salt-and-pepper, Gaussian or Poisson feature perturbation in FPC (ablation)FPC · OOD DomainNet, ResNet-50accuracyImproved
use generator-based synthesized noise for FPC perturbation (ablation)FPC · OOD DomainNet, ResNet-50accuracyHurt
apply FGS under dual noise (pretraining and downstream label noise)ResNet-50 YFCC15M · CIFAR-100 with downstream noise 0-50%accuracy: drop 12-23% smaller than NMTune at 30% pretraining noiseImproved
increase pretraining label noise ratioResNet-50 linear probe · 12 downstream tasksaccuracy vs principal directional angleHurt

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