Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain transfer through deep activation matching
Haoshuo Huang, Qixing Huang, Philipp Krahenbuhl
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 weight-drift regularization to GAM (ablation) | GAM · SVHN->MNIST | accuracy: 73.6 -> 74.6 | Improved |
| add weight-drift regularization to GAM (ablation) | GAM · MNIST->USPS | accuracy: 95.6 -> 95.7 | No clear effect |
| add our regularization to ADDA | ADDA · SVHN->MNIST | accuracy: 72.0 -> 73.5 | Improved |
| add our regularization to ADDA | ADDA · GTA->Cityscapes | IoU: 28.6 -> 30.5 | Improved |
| add Gaussian i.i.d. activation matching | ERFNet · GTA->Cityscapes | IoU: 15.8 -> 28.0 | Improved |
| replace Gaussian matching with GAN-based general activation matching (ablation) | GAM · GTA->Cityscapes | IoU: 28.0 -> 31.3 | Improved |
| replace ADDA with GAM | ERFNet · GTA->Cityscapes | IoU: 28.6 -> 31.3 | Improved |
| replace CyCADA with GAM | Dilated Residual Network · GTA->Cityscapes | IoU: 39.5 -> 40.2 | Improved |
| adapt SYNTHIA-trained model with GAM | ERFNet · SYNTHIA->Cityscapes | IoU: 45.19 -> 62.26 | Improved |
| finetune GAM from Curriculum model | Curriculum · SYNTHIA->Cityscapes | IoU: 29.0 -> 30.7 | Improved |
| skew target label distribution (full SVHN source, odd MNIST target) | GAM · SVHN->MNIST | accuracy: 50.8 -> 29.8 | Hurt |
| adapt with GAM when source skewed (odd SVHN to full MNIST) | GAM · SVHN->MNIST | accuracy: 32.5 -> 36.9 | Improved |
Closest papers by meaning
- An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation (AAAI 2020 · similarity 0.92)
- Learning From Synthetic Data: Addressing Domain Shift for Semantic Segmentation (CVPR 2018 · similarity 0.92)
- Category-Level Adversarial Adaptation for Semantic Segmentation using Purified Features (TPAMI 2021 · similarity 0.92)
- Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation (CVPR 2019 · similarity 0.91)
- Label-Driven Reconstruction for Domain Adaptation in Semantic Segmentation (ECCV 2020 · similarity 0.91)
- CrDoCo: Pixel-Level Domain Transfer With Cross-Domain Consistency (CVPR 2019 · similarity 0.91)
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.
