AI Paper Atlas

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

SUMix: Mixup with Semantic and Uncertain Information

Huafeng Qin, Xin Jin, Hongyu Zhu, Hongchao Liao, Mounim A. El Yacoubi, Xinbo Gao

ECCV 2024 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 9 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 SUMix (learned mix ratio plus uncertainty regularizer) to cutting-based mixupCutMix, FMix, SaliencyMix, ResizeMix · CIFAR-100 ResNet18top-1 accuracy: average +0.82Improved
add SUMix to CutMixCutMix on W-ResNet28-8 · CIFAR-100top-1 accuracy: 84.45 -> 84.56No clear effect
add SUMix to cutting-based mixupCutMix, FMix, SaliencyMix, ResizeMix · Tiny-ImageNet ResNeXt50top-1 accuracy: average +2.07Improved
add SUMix to cutting-based mixupCutMix, FMix, SaliencyMix, ResizeMix · ImageNet-1K ResNet18top-1 accuracy: average +0.47Improved
add SUMix to cutting-based mixupFMix, SaliencyMix, ResizeMix, CutMix on Swin-Tiny · CIFAR-100 Swin-Tinytop-1 accuracy: average +0.23; FMix 80.72 -> 80.73No clear effect
add SUMix to saliency-based mixupPuzzleMix, AutoMix · CIFAR-100 ResNeXt50top-1 accuracy: AutoMix 82.84 -> 83.82Improved
add SUMix to cutting-based mixupCutMix, FMix, SaliencyMix, ResizeMix · CIFAR-100-Ccorruption accuracy: CutMix 43.06 -> 44.31Improved
add SUMix to mixupCutMix, FMix, SaliencyMix, ResizeMix · CIFAR-100 FGSM attackFGSM error: CutMix 91.15 -> 90.41; SaliencyMix 89.64 -> 91.49Mixed
add redefined per-sample mix ratio (ablation)CutMix · CIFAR-100 ResNet18top-1 accuracy: 78.17 -> 79.13Improved
add semantic-only regularization loss (ablation)SaliencyMix · CIFAR-100 ResNet18top-1 accuracy: 79.12 -> 78.71Hurt
add uncertainty-only regularization loss (ablation)CutMix · CIFAR-100 ResNet18top-1 accuracy: 78.17 -> 79.60Improved
combine semantic and uncertainty regularization (ablation)CutMix · CIFAR-100 ResNet18top-1 accuracy: 79.60 -> 79.78Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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