Research map / Label-efficient and robust learning / Domain adaptation and generalization
TRUST: Leveraging Text Robustness for Unsupervised Domain Adaptation
Mattia Litrico, Mario Valerio Giuffrida, Sebastiano Battiato, Devis Tuia
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 language-guided soft-contrastive loss to pseudo-label training (ablation) | pseudo-labels only · GeoNet | avg accuracy: 62.31 -> 65.38 | Improved |
| add standard hard contrastive loss to pseudo-label training (ablation) | pseudo-labels only · GeoNet | avg accuracy: 62.31 -> 62.86 | No clear effect |
| add CLIP-based uncertainty estimation to pseudo-label training (ablation) | pseudo-labels only · GeoNet | avg accuracy: 62.31 -> 63.41 | Improved |
| add CLIP uncertainty on top of soft-contrastive loss (ablation) | TRUST · GeoNet | avg accuracy: 65.38 -> 66.40 | Improved |
| generate pseudo-labels with source-finetuned BERT instead of CLIP zero-shot (ablation) | TRUST · GeoImnet | accuracy: 61.36 -> 66.40 | Improved |
| generate pseudo-labels with BERT instead of source-finetuned CLIP (ablation) | TRUST · GeoImnet | accuracy: 63.32 -> 66.40 | Improved |
| generate pseudo-labels with BERT instead of CLIP-Text (ablation) | TRUST · GeoImnet | accuracy: 64.11 -> 66.40 | Improved |
| use TRUST instead of LaGTran | LaGTran · DomainNet-345 (average) | accuracy: 72.41 -> 74.95 | Improved |
| use TRUST instead of LaGTran | LaGTran · GeoNet (total average) | accuracy: 60.24 -> 63.20 | Improved |
| use TRUST instead of PADCLIP | PADCLIP · VisDA | accuracy: 88.5 -> 90.6 | Improved |
| use TRUST in non-real to real transfers on DomainNet versus CLIP | CLIP · DomainNet-345 (non-real to real) | accuracy | Mixed |
Closest papers by meaning
- Toward Improving Robustness and Accuracy in Unsupervised Domain Adaptation (AAAI 2025 · similarity 0.88)
- Split to Merge: Unifying Separated Modalities for Unsupervised Domain Adaptation (CVPR 2024 · similarity 0.88)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.88)
- Target Semantics Clustering via Text Representations for Robust Universal Domain Adaptation (AAAI 2025 · similarity 0.87)
- UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models (NeurIPS 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.
