Research map / Label-efficient and robust learning / Self-supervised and contrastive
Domain Invariant Masked Autoencoders for Self-Supervised Learning from Multi-Domains
Haiyang Yang, Shixiang Tang, Meilin Chen, Yizhou Wang, Feng Zhu, Lei Bai, Rui Zhao, Wanli Ouyang
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 content-preserved CP-StyleMix instead of no augmentation (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 36.85 -> 48.56 | Improved |
| use CP-StyleMix instead of Mixup (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 38.09 -> 48.56 | Improved |
| use CP-StyleCut instead of CutMix (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 37.50 -> 47.21 | Improved |
| add content-mix Mixup/CutMix instead of no augmentation (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 36.85 -> 38.09 | No clear effect |
| add Mixup mixing step after style transfer (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 46.11 -> 48.56 | Improved |
| add CutMix mixing step after style transfer (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 46.11 -> 47.21 | Improved |
| add multiple domain-specific decoders instead of single decoder (CP-StyleMix) (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 38.09 -> 48.56 | Improved |
| add multiple domain-specific decoders instead of single decoder (CP-StyleCut) (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 37.50 -> 47.21 | Improved |
| increase decoder depth from 1 to 8 (multi decoders) (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 44.93 -> 48.56 | Improved |
| increase decoder depth from 1 to 8 (single decoder) (ablation) | MAE-style single decoder · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 37.46 -> 38.09 | No clear effect |
| increase decoder depth from 8 to 12 (multi decoders) (ablation) | DiMAE · DomainNet Clipart linear eval (ViT-Tiny, 100 epochs) | top-1 accuracy: 48.56 -> 46.11 | Hurt |
| use DiMAE instead of MAE for in-domain pretraining | ViT-small · PACS in-domain (linear eval) | avg accuracy: 71.2 -> 72.0 | Improved |
Closest papers by meaning
- Cycle-consistent Masked AutoEncoder for Unsupervised Domain Generalization (ICLR 2023 · similarity 0.89)
- Reconstruction Target Matters in Masked Image Modeling for Cross-Domain Few-Shot Learning (AAAI 2025 · similarity 0.88)
- Unsupervised Domain Generalization by Learning a Bridge Across Domains (CVPR 2022 · similarity 0.88)
- Disentangling Masked Autoencoders for Unsupervised Domain Generalization (ECCV 2024 · similarity 0.88)
- Self-Guided Masked Autoencoders for Domain-Agnostic Self-Supervised Learning (ICLR 2024 · similarity 0.86)
- Towards domain-invariant Self-Supervised Learning with Batch Styles Standardization (ICLR 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.
