Research map / Label-efficient and robust learning / Domain adaptation and generalization
HVCLIP: High-dimensional Vector in CLIP for Unsupervised Domain Adaptation
Noranart Vesdapunt, Kah Kuen Fu, Yue Wu, Xu Zhang, Pradeep Natarajan
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 |
|---|---|---|---|
| naively fine-tune CLIP on UDA data instead of zero-shot (ablation) | CLIP (ResNet-101) · VisDA-2017 | accuracy (%): 82.3 -> 57.1 | Hurt |
| add Forgetting Reduction in hypervector space (dim 4096) (ablation) | HVCLIP · VisDA-2017 | accuracy (%): 57.1 -> 86.5 | Improved |
| add Discrepancy Reduction (ablation) | HVCLIP · VisDA-2017 | accuracy (%): 86.5 -> 87.6 | Improved |
| add Feature Augmentation (ablation) | HVCLIP · VisDA-2017 | accuracy (%): 87.6 -> 88.1 | Improved |
| add cycle loss to Feature Augmentation (ablation) | HVCLIP · VisDA-2017 | accuracy (%): 88.1 -> 89.1 | Improved |
| convert features to hypervector space (ablation) | HVCLIP · VisDA-2017 | accuracy (%): 68.5 -> 89.1 | Improved |
| increase hypervector dimension from 64 to 4096 (ablation) | HVCLIP · VisDA-2017 | accuracy (%): 68.4 -> 86.5 | Improved |
| use Forgetting Reduction instead of PADCLIP CFM | HVCLIP · VisDA-2017 | accuracy (%): 86.0 -> 86.5 | Improved |
| combine PADCLIP CFM with Discrepancy Reduction and Feature Augmentation (ablation) | PADCLIP · VisDA-2017 | accuracy (%): 86.0 -> 72.1 | Hurt |
| apply HVCLIP in continual learning versus ZSCL | HVCLIP · CIFAR-100 continual learning, 20 steps (last) | accuracy (%): 69.6 -> 72.0 | Improved |
| apply HVCLIP to BiT and DeiT backbones versus PADCLIP | HVCLIP · VisDA-2017 (DeiT) | accuracy (%): 88.4 -> 89.9 | Improved |
| add Feature Augmentation training cost | HVCLIP · VisDA-2017 training time | training hours: 23 -> 30 hours total | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FixBi — FixBi: Bridging Domain Spaces for Unsupervised Domain Adaptation (CVPR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation (CVPR 2025 · similarity 0.87)
- Learning to Adapt Frozen CLIP for Few-Shot Test-Time Domain Adaptation (ICLR 2025 · similarity 0.87)
- CLIPCEIL: Domain Generalization through CLIP via Channel rEfinement and Image-text aLignment (NeurIPS 2024 · similarity 0.87)
- Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · similarity 0.86)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.86)
- un$^2$CLIP: Improving CLIP's Visual Detail Capturing Ability via Inverting unCLIP (NeurIPS 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.
