Research map / Vision-language and multimodal / CLIP and zero-shot
Improving CLIP Adaptation by Breaking Tail Alignment for Source-Free Cross-Domain Few-Shot Learning
Shuai Yi, Yixiong Zou, Yuhua Li, Ruixuan Li
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 push-away of tail tokens (ablation) | CLIP-LoRA · CDFSL 5-way 5-shot avg | accuracy: 65.99 -> 67.45 | Improved |
| add head-token enhancement on top of tail repulsion (ablation) | Push-away tail · CDFSL 5-way 5-shot avg | accuracy: 67.45 -> 68.53 | Improved |
| use direct feature modulation instead of loss constraint (ablation) | ATHA · CDFSL 5-way 5-shot avg | accuracy: 66.76 -> 68.53 | Improved |
| use MaxSim/MinSim text instead of weighted average text (ablation) | ATHA · CDFSL 5-way 5-shot avg | accuracy: 67.73 -> 68.53 | Improved |
| make alpha, beta learnable per layer instead of fixed (ablation) | ATHA · CDFSL 5-way 5-shot avg | accuracy: 68.26 -> 68.53 | Improved |
| add PACL alignment loss to LoRA | CLIP-LoRA · CDFSL 5-way 5-shot avg | accuracy: 65.99 -> 65.85 | Hurt |
| add SPARC alignment to LoRA | CLIP-LoRA · CDFSL 5-way 5-shot avg | accuracy: 65.99 -> 65.55 | Hurt |
| replace cosine similarity with Euclidean distance for token selection (ablation) | ATHA · CDFSL 5-way 5-shot avg | accuracy: 68.53 -> 68.39 | No clear effect |
| add ATHA to CLIP ViT-L/14 LoRA | CLIP ViT-L/14 · CDFSL 5-way 1-shot avg | accuracy: 54.90 -> 56.90 | Improved |
| add ATHA to SigLIP LoRA | SigLIP · CDFSL 5-way 1-shot avg | accuracy: 49.83 -> 53.11 | Improved |
| add ATHA to PEcore LoRA | PEcore · CDFSL 5-way 1-shot avg | accuracy: 56.61 -> 58.29 | Improved |
| add ATHA to CoOp | CoOp · CDFSL 5-way 5-shot avg | accuracy: 60.29 -> 62.03 | Improved |
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.
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · 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
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.90)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.89)
- Unlocking Fine-Grained Perception in CLIP via Structurally-Aware Latent Masked Modeling (NeurIPS 2026 · similarity 0.89)
- Kernel-based Unsupervised Embedding Alignment for Enhanced Visual Representation in Vision-language Models (ICML 2025 · similarity 0.89)
- FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs (CVPR 2025 · similarity 0.89)
- Enhancing CLIP Robustness via Cross-Modality Alignment (NeurIPS 2025 · 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.
