Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Domain Generalization by Learning a Bridge Across Domains
Check 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 a learned edge-regularized bridge domain to self-supervised contrastive training (ablation) | MoCo V2 for DomainNet FUDA · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): +13.7/+14.2 | Helped |
| Add multiple domain-specific negative queues (ablation) | MoCo V2 for DomainNet FUDA · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): +4.2/+0.4 | Helped |
| Add both multiple negative queues and a domain discriminator (ablation) | MoCo V2 for DomainNet FUDA · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): +3.8/+6.0 | Helped |
| Add a domain discriminator alone (ablation) | MoCo V2 for DomainNet FUDA · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): −0.3/+0.1 | Mixed |
| Use fixed Canny edge maps as BrAD mappings (ablation) | DD + MQ contrastive model · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): +10.9/+11.9 | Helped |
| Replace Canny BrAD with frozen pretrained HED (ablation) | DD + MQ contrastive model · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): −1.7/−1.8 | Hurt |
| Learn the HED-based BrAD mappings jointly end-to-end (ablation) | DD + MQ contrastive model · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): +2.8/+2.3 | Helped |
| Use transductive unlabeled-domain data (ablation) | Full BrAD FUDA model · DomainNet FUDA | accuracy (%): ∼ 4% | Helped |
| Use ImageNet pretraining (ablation) | Full BrAD FUDA model · DomainNet FUDA | accuracy (%): ∼ 10% | Helped |
| Remove BSDS500 pretrained weights and use randomly initialized mappings with blurred Canny supervision (ablation) | Full BrAD FUDA model · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): −1.2/+1.0 | Mixed |
| Use learned BrAD rather than no BrAD under ImageNet pretraining and transductive training (ablation) | Full BrAD FUDA model · DomainNet FUDA, 1-shot and 3-shot | accuracy (%): +9.9/+7.2 | Helped |
| Replace HED-based L2 BrAD loss with Canny-based L2 loss (ablation) | Learned BrAD FUDA model · DomainNet FUDA, 1-shot | accuracy (%): 48.64 -> 47.40 | Hurt |
Closest papers by meaning
- Multimodal Unsupervised Domain Generalization by Retrieving Across the Modality Gap (ICLR 2025 · similarity 0.88)
- Minimal Semantic Sufficiency Meets Unsupervised Domain Generalization (NeurIPS 2025 · similarity 0.88)
- START: A Generalized State Space Model with Saliency-Driven Token-Aware Transformation (NeurIPS 2024 · similarity 0.87)
- DGFamba: Learning Flow Factorized State Space for Visual Domain Generalization (AAAI 2025 · similarity 0.87)
- Towards domain-invariant Self-Supervised Learning with Batch Styles Standardization (ICLR 2024 · similarity 0.87)
- Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation (CVPR 2025 · 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.