Research map / Label-efficient and robust learning / Domain adaptation and generalization
Semantic-guided Robustness Tuning for Few-Shot Transfer Across Extreme Domain Shift
kangyu xiao, Zilei Wang, junjie 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 SRT fine-tuning to CLIP | CLIP ViT-B · BSCD-FSL 1-shot avg cls-I | accuracy: 50.93 -> 57.12 | Improved |
| traditional fine-tuning of the last block on DINO | DINO ViT-B · BSCD-FSL 1-shot avg cls-I | accuracy: -14.07 | Hurt |
| full fine-tuning CLIP with Trad-FT | CLIP ViT-B · BSCD-FSL 1-shot avg cls-I | accuracy: -5.31 | Hurt |
| add L_RI loss (ablation) | SRT on CLIP · EuroSAT 1-shot | accuracy: 68.28 -> 73.72 | Improved |
| add L_CLIP on top of L_RI (ablation) | SRT on CLIP · EuroSAT 1-shot | accuracy: 73.72 -> 81.03 | Improved |
| add semantic injection via MMIT-Mixup versus none (ablation) | SRT on CLIP · EuroSAT 1-shot cls-T | accuracy: 65.28 -> 74.95 | Improved |
| replace MMIT-Mixup with cross-attention injection on DINO (ablation) | SRT on DINO · ISIC 1-shot cls-I | accuracy: 35.64 -> 32.98 | Hurt |
| cross-attention injection versus none on CLIP (ablation) | SRT on CLIP · EuroSAT 1-shot cls-I | accuracy: 72.45 -> 73.29 | Improved |
| MMIT-Mixup versus cross-attention in extra resources (ablation) | SRT · FLOPs and params | extra parameters: 2.16 M -> 0 | Improved |
| apply SRT to lightweight DINO ViT-S | DINO ViT-S · BSCD-FSL 1-shot avg cls-I | accuracy: 51.98 -> 55.33 | Improved |
Closest papers by meaning
- Lipsum-FT: Robust Fine-Tuning of Zero-Shot Models Using Random Text Guidance (ICLR 2024 · similarity 0.88)
- Reconstruction Target Matters in Masked Image Modeling for Cross-Domain Few-Shot Learning (AAAI 2025 · similarity 0.87)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.86)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.86)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.86)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 2025 · similarity 0.86)
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.
