AI Paper Atlas

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

ECCV 2024 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Geometric alignment and synthetic-to-real adaptation · 3 citations (OpenAlex)

Read the paper

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 changedWhereMeasuredEffect
naively fine-tune CLIP on UDA data instead of zero-shot (ablation)CLIP (ResNet-101) · VisDA-2017accuracy (%): 82.3 -> 57.1Hurt
add Forgetting Reduction in hypervector space (dim 4096) (ablation)HVCLIP · VisDA-2017accuracy (%): 57.1 -> 86.5Improved
add Discrepancy Reduction (ablation)HVCLIP · VisDA-2017accuracy (%): 86.5 -> 87.6Improved
add Feature Augmentation (ablation)HVCLIP · VisDA-2017accuracy (%): 87.6 -> 88.1Improved
add cycle loss to Feature Augmentation (ablation)HVCLIP · VisDA-2017accuracy (%): 88.1 -> 89.1Improved
convert features to hypervector space (ablation)HVCLIP · VisDA-2017accuracy (%): 68.5 -> 89.1Improved
increase hypervector dimension from 64 to 4096 (ablation)HVCLIP · VisDA-2017accuracy (%): 68.4 -> 86.5Improved
use Forgetting Reduction instead of PADCLIP CFMHVCLIP · VisDA-2017accuracy (%): 86.0 -> 86.5Improved
combine PADCLIP CFM with Discrepancy Reduction and Feature Augmentation (ablation)PADCLIP · VisDA-2017accuracy (%): 86.0 -> 72.1Hurt
apply HVCLIP in continual learning versus ZSCLHVCLIP · CIFAR-100 continual learning, 20 steps (last)accuracy (%): 69.6 -> 72.0Improved
apply HVCLIP to BiT and DeiT backbones versus PADCLIPHVCLIP · VisDA-2017 (DeiT)accuracy (%): 88.4 -> 89.9Improved
add Feature Augmentation training costHVCLIP · VisDA-2017 training timetraining hours: 23 -> 30 hours totalHurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy