Research map / Vision-language and multimodal / CLIP and zero-shot
SuS-X: Training-Free Name-Only Transfer of Vision-Language Models
Vishaal Udandarao, Ankush Gupta, Samuel Albanie
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 |
|---|---|---|---|
| replace TIP-Adapter with TIP-X when using the LAION support set (ablation) | SuS-X-LC · 19-dataset average (default prompts) | accuracy: 53.83 -> 54.20 | Improved |
| replace TIP-Adapter with TIP-X when using the SD support set (ablation) | SuS-X-SD · 19-dataset average (CuPL+e prompts) | accuracy: 56.63 -> 56.73 | Improved |
| add curated support set (SuS) to zero-shot CLIP (ablation) | Zero-shot CLIP · 19-dataset average | accuracy: 52.27 -> 53.49 | Improved |
| use CuPL prompting instead of Photo for SD support set generation (ablation) | SuS-X-SD · 19-dataset average | accuracy: 56.32 -> 56.73 | Improved |
| use CuPL prompting instead of Photo for LAION retrieval (ablation) | SuS-X-LC · 19-dataset average | accuracy: 56.87 -> 56.20 | Hurt |
| CuPL prompting diversifies support set images (ablation) | SuS · ImageNet | diversity: 0.28 -> 0.32 (LC), 0.17 -> 0.20 (SD) | Improved |
| increase support set size (ablation) | SuS-X · per-dataset | accuracy | Mixed |
| sweep gamma from 0 to 1 in TIP-X (ablation) | TIP-X · DTD | accuracy: 47.16 -> 47.96 | Improved |
| use TIP-X instead of TIP-Adapter in the few-shot training-free regime | TIP-X · 11-dataset few-shot (1-16 shots) | accuracy: +0.91% absolute gain across shots | Improved |
| apply SuS-X to other VLMs (TCL) | TCL · 19-dataset average | accuracy: 31.38 -> 42.75 | Improved |
Closest papers by meaning
- Reproducible Vision-Language Models Meet Concepts Out of Pre-Training (CVPR 2025 · similarity 0.90)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.90)
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 2024 · similarity 0.90)
- ProKeR: A Kernel Perspective on Few-Shot Adaptation of Large Vision-Language Models (CVPR 2025 · similarity 0.89)
- Dual Memory Networks: A Versatile Adaptation Approach for Vision-Language Models (CVPR 2024 · similarity 0.89)
- UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models (NeurIPS 2024 · similarity 0.88)
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.
