Research map / Label-efficient and robust learning / Recognition and long-tail
AutoMix: Unveiling the Power of Mixup for Stronger Classifiers
Zicheng Liu, Siyuan Li, Di Wu, Zihan Liu, Zhiyuan Chen, Lirong Wu, Stan Z. Li
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-attention, lambda embedding and lambda loss to MixBlock (ablation) | AutoMix · Tiny-ImageNet | top-1 accuracy: 66.83 -> 70.72 | Improved |
| add EMA and CE to the momentum pipeline of MixBlock (ablation) | AutoMix · CIFAR-100 and Tiny-ImageNet | top-1 accuracy | Improved |
| MixBlock without EMA and CE versus vanilla training (ablation) | MixBlock · CIFAR-100 | top-1 accuracy: +2.29% | Improved |
| MixBlock without EMA and CE versus vanilla training (ablation) | MixBlock · Tiny-ImageNet | top-1 accuracy: +2.21% | Improved |
| add CE loss to other mixup methods (ablation) | other mixup methods · CIFAR-100 | top-1 accuracy | No clear effect |
| set mixup alpha to 2 instead of other values (ablation) | AutoMix · CIFAR-100 and Tiny-ImageNet | top-1 accuracy | Improved |
| use feature layer l=3 for mixing (ablation) | AutoMix · Tiny-ImageNet | top-1 accuracy | Improved |
| use AutoMix instead of AutoMix-free ResNet-18 baseline on ImageNet 100 epochs (ablation) | ResNet-18 · ImageNet-1k | top-1 accuracy: 70.04 -> 70.50 | Improved |
| use AutoMix instead of vanilla on EfficientNet B0 | EfficientNet B0 · ImageNet-1k | top-1 accuracy: +0.32% | Improved |
| use CutMix with longer training on CIFAR-100 | CutMix · CIFAR-100 | top-1 accuracy: 79.58 -> 78.29 | Hurt |
| use AutoMix instead of PuzzleMix on Tiny-ImageNet ResNeXt-50 | ResNeXt-50 · Tiny-ImageNet | top-1 accuracy: 67.83 -> 70.72 | Improved |
| use AutoMix instead of vanilla on CUB-200 | ResNet-18 · CUB-200 | top-1 accuracy: 77.68 -> 79.87 | Improved |
Closest papers by meaning
- Adversarial AutoMixup (ICLR 2024 · similarity 0.94)
- AutoMix: Mixup Networks for Sample Interpolation via Cooperative Barycenter Learning (ECCV 2020 · similarity 0.91)
- SuperMix: Supervising the Mixing Data Augmentation (CVPR 2021 · similarity 0.89)
- Harnessing Hard Mixed Samples with Decoupled Regularizer (NeurIPS 2023 · similarity 0.89)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.87)
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · similarity 0.86)
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.
