Research map / Label-efficient and robust learning / Domain adaptation and generalization
Self-ensembling for visual domain adaptation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 confidence thresholding in place of Gaussian ramp-up (ablation) | Mean teacher domain adaptation · STL → CIFAR-10 | accuracy: 18.3 -> 71.65 | Helped |
| Add confidence thresholding in place of Gaussian ramp-up (ablation) | Mean teacher domain adaptation · MNIST → SVHN | accuracy: 13.96 -> 33.87 | Helped |
| Add confidence thresholding (ablation) | Mean teacher domain adaptation · MNIST ↔ USPS | accuracy: 98.07 -> 92.35; 98.26 -> 88.14 | Mixed |
| Add translations and horizontal flips (ablation) | Mean teacher domain adaptation · CIFAR-10 ↔ STL | accuracy: 77.53 -> 79.73; 71.65 -> 74.24 | Mixed |
| Add affine augmentation (ablation) | Mean teacher with confidence thresholding and standard augmentation · STL → CIFAR-10 | accuracy: 74.24 -> 69.86 | Hurt |
| Add affine augmentation (ablation) | Mean teacher with confidence thresholding and standard augmentation · MNIST → SVHN | accuracy: 34.15 -> 37.49 | Helped |
| Add class balance loss (ablation) | Mean teacher domain adaptation · MNIST → SVHN | accuracy: 37.49 -> 97.0 | Helped |
| Use MNIST-specific intensity augmentation (ablation) | Mean teacher with class balance loss · MNIST → SVHN | accuracy: 37.49 -> 97.0 | Helped |
| Use competition-specific augmentation instead of reduced augmentation (ablation) | ResNet-152 domain adaptation · VisDA-17 validation | mean class accuracy: 85.4 -> 84.29 | Hurt |
| Add test-time augmentation to reduced augmentation (ablation) | ResNet-152 domain adaptation · VisDA-17 validation | mean class accuracy: 85.4 -> 86.6 | Helped |
| Add test-time augmentation to reduced augmentation (ablation) | ResNet-152 domain adaptation · VisDA-17 test | mean class accuracy: 91.17 -> 92.25 | Helped |
| Use translations and horizontal flips (ablation) | Supervised source training · MNIST → USPS | accuracy: 77.55 -> 77.53 | No clear effect |
Closest papers by meaning
- Visual Domain Adaptation by Consensus-Based Transfer to Intermediate Domain (AAAI 2020 · similarity 0.89)
- Learning an Invariant Hilbert Space for Domain Adaptation (CVPR 2017 · similarity 0.89)
- Self-Ensembling Attention Networks: Addressing Domain Shift for Semantic Segmentation (AAAI 2019 · similarity 0.88)
- AnyDA: Anytime Domain Adaptation (ICLR 2023 · similarity 0.88)
- Exploring Category-Agnostic Clusters for Open-Set Domain Adaptation (CVPR 2020 · similarity 0.88)
- Heuristic Domain Adaptation (NeurIPS 2020 · similarity 0.87)
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.