Research map / Label-efficient and robust learning / Domain adaptation and generalization
Drop to Adapt: Learning Discriminative Features for Unsupervised Domain Adaptation
Seungmin Lee, Dongwan Kim, Namil Kim, Seong-Gyun Jeong
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 VAT to cDTA (ablation) | DTA (ResNet-101) · VisDA-2017 classification | average accuracy (%): 77.0 -> 81.2 | Improved |
| add VAT to fDTA (ablation) | DTA (ResNet-101) · VisDA-2017 classification | average accuracy (%): 76.4 -> 80.5 | Improved |
| add VAT to fDTA (ablation) | DTA (ResNet-50) · VisDA-2017 classification | average accuracy (%): 72.5 -> 73.1 | Improved |
| add VAT to cDTA (ablation) | DTA (ResNet-50) · VisDA-2017 classification | average accuracy (%): 74.7 -> 75.4 | Improved |
| add fDTA to cDTA + VAT (full DTA) (ablation) | DTA (ResNet-101) · VisDA-2017 classification | average accuracy (%): 81.2 -> 81.5 | Improved |
| add fDTA to cDTA + VAT (full DTA) (ablation) | DTA (ResNet-50) · VisDA-2017 classification | average accuracy (%): 75.4 -> 76.2 | Improved |
| add VAT to cDTA + fDTA (ablation) | DTA (ResNet-101) · VisDA-2017 classification | average accuracy (%): 77.4 -> 81.5 | Improved |
| add fDTA to cDTA (ablation) | DTA (ResNet-50) · VisDA-2017 classification | average accuracy (%): 74.7 -> 75.0 | No clear effect |
| add the full DTA objective to source-only training | ResNet-101 · VisDA-2017 classification | average accuracy (%): 50.8 -> 81.5 | Improved |
| add DTA to source-only training | ResNet-50 FCN · GTA5 to Cityscapes | mIoU (%): 24.8 -> 35.8 | Improved |
| use DTA instead of ADR | ResNet-50 FCN · GTA5 to Cityscapes | mIoU (%): 33.3 -> 35.8 | Improved |
| use DTA when STL is the (small) source domain | DTA · STL to CIFAR | target accuracy (%): 74.2 -> 72.8 | Hurt |
Closest papers by meaning
- Adversarial Dropout Regularization (ICLR 2018 · similarity 0.91)
- Attending to Discriminative Certainty for Domain Adaptation (CVPR 2019 · similarity 0.90)
- Adversarial Discriminative Domain Adaptation (CVPR 2017 · similarity 0.90)
- A DIRT-T Approach to Unsupervised Domain Adaptation (ICLR 2018 · similarity 0.90)
- Mind the Discriminability: Asymmetric Adversarial Domain Adaptation (ECCV 2020 · similarity 0.89)
- Bridging Theory and Algorithm for Domain Adaptation (ICML 2019 · similarity 0.89)
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.
