Research map / Label-efficient and robust learning / Domain adaptation and generalization
Universal Cross-Domain Retrieval: Generalizing Across Classes and Domains
Soumava Paul, Titir Dutta, Soma Biswas
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 semantic neighbourhood loss (kappa=2) (ablation) | Base N/W · Sketchy extended ZS-SBIR (ablation) | mAP@200: 0.5218 -> 0.5613 | Improved |
| use class-relative weighting kappa=2 instead of equal weights kappa=0 in semantic neighbourhood loss (ablation) | Base N/W + L_Sn · Sketchy extended ZS-SBIR (ablation) | mAP@200: 0.5593 -> 0.5613 | Improved |
| add mixup cross-entropy alone (ablation) | Base N/W · Sketchy extended ZS-SBIR (ablation) | mAP@200: 0.5218 -> 0.5252 | No clear effect |
| add mixture prediction loss on top of mixup cross-entropy (ablation) | Base N/W + L_CE^mix · Sketchy extended ZS-SBIR (ablation) | mAP@200: 0.5252 -> 0.5665 | Improved |
| add semantic neighbourhood loss on top of mixup and mixture prediction losses (ablation) | Base N/W + L_CE^mix + L_Mp · Sketchy extended ZS-SBIR (ablation) | mAP@200: 0.5665 -> 0.5781 | Improved |
| add both losses and mixing to the base network (full SnMpNet) (ablation) | Base N/W · Sketchy extended ZS-SBIR (ablation) | Prec@200: 0.4497 -> 0.5155 | Improved |
| remove the domain indicator from SAKE-512 | SAKE-512 · Sketchy extended ZS-SBIR | mAP@200: 0.6246 -> 0.5484 | Hurt |
| replace two-branch Doodle-to-Search with a single branch | Doodle-to-Search · Sketchy extended ZS-SBIR | mAP@200: 0.4606 -> 0.3743 | Hurt |
| remove domain-discriminator loss from Doodle-SingleNet | Doodle-SingleNet · Sketchy extended ZS-SBIR | mAP@200: 0.3743 -> 0.3726 | No clear effect |
| replace VGG-16 with SE-ResNet50 backbone | Doodle-SingleNet · Sketchy extended ZS-SBIR | mAP@200: 0.3743 -> 0.4022 | Improved |
| reduce SAKE-w/o Label embedding from 512 to 300 dims | SAKE-w/o Label · Sketchy extended ZS-SBIR | mAP@200: 0.5484 -> 0.5192 | Hurt |
| train with five domains instead of Sketch and Real only (QuickDraw query, unseen classes) (ablation) | SnMpNet · DomainNet UCDR, QuickDraw query | mAP@200: 0.1540 -> 0.1736 | Improved |
Closest papers by meaning
- Semantic Feature Learning for Universal Unsupervised Cross-Domain Retrieval (NeurIPS 2024 · similarity 0.90)
- Retrieval Across Any Domains via Large-scale Pre-trained Model (ICML 2024 · similarity 0.89)
- Unsupervised Cross-Domain Image Retrieval via Prototypical Optimal Transport (AAAI 2024 · similarity 0.86)
- NeighborRetr: Balancing Hub Centrality in Cross-Modal Retrieval (CVPR 2025 · similarity 0.86)
- Towards Generalizing to Unseen Domains with Few Labels (CVPR 2024 · similarity 0.86)
- Tackling Dimensional Collapse toward Comprehensive Universal Domain Adaptation (ICML 2025 · similarity 0.85)
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.
