Research map / Vision-language and multimodal / Cross-modal learning
Mind the Discriminability Trap in Source-Free Cross-domain Few-shot Learning
Zhenyu Zhang, Yixiong Zou, Yuhua Li, Ruixuan Li, Guangyao Chen
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 SVL (suppress visual learning) (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 56.07 -> 57.55 | Improved |
| add RA (relationship alignment) (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 56.07 -> 57.17 | Improved |
| add both SVL and RA (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 56.07 -> 58.26 | Improved |
| add RA on top of SVL (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 57.55 -> 58.26 | Improved |
| apply L_ad only at the beginning of fine-tuning (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 56.07 -> 57.55 | Improved |
| apply L_ad only in the last epochs (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 56.07 -> 55.99 | No clear effect |
| apply L_ad in the middle epochs (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 56.07 -> 57.04 | Improved |
| add visual-learning loss L_v (ablation) | CLIP fine-tune · ChestX 5-way 5-shot | modality loss gap: 0.356 -> 0.742 | Hurt |
| add L_ad to fine-tuning (ablation) | CLIP fine-tune · ChestX 5-way 5-shot | modality loss gap: 0.356 -> 0.249 | Improved |
| add L_ad to fine-tuning (ablation) | PE-Core fine-tune · ISIC 5-way 5-shot | modality loss gap: 0.538 -> 0.254 | Improved |
| add the method to CoOp (ablation) | CoOp · CDFSL 5-way 1-shot avg | accuracy: 51.77 -> 52.36 | Improved |
| add the method to Maple (ablation) | Maple · CDFSL 5-way 1-shot avg | accuracy: 53.07 -> 54.05 | 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 2 of 2 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
- A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.89)
- Random Registers for Cross-Domain Few-Shot Learning (ICML 2025 · similarity 0.89)
- Attention Temperature Matters in ViT-Based Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.89)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.88)
- CF-VLM:CounterFactual Vision-Language Fine-tuning (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.
