Research map / Vision-language and multimodal / CLIP and zero-shot
Beyond Retraining: Training-Free Unknown Class Filtering for Source-Free Open Set Domain Adaptation of Vision–Language Models
Yongguang Li, Jindong Li, Qi Wang, QianLi Xing, Runliang Niu, Shengsheng Wang, Menglin Yang
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 SUFF unknown-subspace filtering to CLIP (ablation) | CLIP zero-shot with threshold · Office-Home + VisDA, avg over thresholds | HOS (%): 72.4 -> 80.1 | Improved |
| add SUFF to SigLIP (ablation) | SigLIP zero-shot · Office-Home + VisDA, avg over thresholds | HOS (%): 67.1 -> 73.9 | Improved |
| add SUFF to ALIGN (ablation) | ALIGN zero-shot · Office-Home + VisDA, avg over thresholds | HOS (%): 72.8 -> 77.7 | Improved |
| add SUFF (ablation) | CLIP · VTAB-6, Office-Home, VisDA | AR@1: 0.46 -> 0.23 | Improved |
| use BGAT instead of fixed threshold (CLIP, no SUFF) (ablation) | CLIP · Office-Home + VisDA avg | HOS (%): 66.9 -> 76.4 | Improved |
| use BGAT instead of fixed threshold (SigLIP, no SUFF) (ablation) | SigLIP · Office-Home + VisDA avg | HOS (%): 52.1 -> 80.0 | Improved |
| use BGAT instead of k-means threshold (CLIP, no SUFF) (ablation) | CLIP · Office-Home + VisDA avg | HOS (%): 74.0 -> 76.4 | Improved |
| add Box-Cox transformation before GMM thresholding (ablation) | BGAT · Office-Home + VisDA, 3 VLMs | HOS (%): avg gain 2.67% | Improved |
| add BGAT to UEO-CLIP | UEO-CLIP · Office-Home (4 domains avg) | HOS (%): 67.0 -> 75.9 | Improved |
| add SUFF on top of BGAT to UEO-CLIP | UEO-CLIP · Office-Home (4 domains avg) | HOS (%): 75.9 -> 78.8 | Improved |
| add BGAT and SUFF to SHOT | SHOT · Office-Home (4 domains avg) | HOS (%): 57.8 -> 62.9 | Improved |
| vary variance retention ratio tau in 0.7-0.9 (ablation) | SUFF · Office-Home + VisDA avg | HOS (%): variance 0.19 (%)^2 | No clear effect |
Closest papers by meaning
- Leveraging Vision-Language Models for Improving Domain Generalization in Image Classification (CVPR 2024 · similarity 0.90)
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 2024 · similarity 0.90)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · similarity 0.89)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · similarity 0.89)
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.
