Research map / Label-efficient and robust learning / Domain adaptation and generalization
Free Lunch to Meet the Gap: Intermediate Domain Reconstruction for Cross-Domain Few-Shot Learning
Tong Zhang, Yifan Zhao, Liangyu Wang, Jia 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 target-domain loss L_tar (ablation) | GNN baseline · EuroSAT 5-shot | accuracy: 83.64 -> 87.59 | Improved |
| add intermediate-proxy loss L_proxy (ablation) | GNN + L_tar · EuroSAT 5-shot | accuracy: 87.59 -> 90.11 | Improved |
| add intermediate alignment loss L_align (ablation) | GNN + L_tar + L_proxy · EuroSAT 5-shot | accuracy: 90.11 -> 91.08 | Improved |
| add L_align (ablation) | GNN + L_tar + L_proxy · EuroSAT 1-shot | accuracy: 68.95 -> 71.60 | Improved |
| add L_proxy (ablation) | GNN + L_tar · Places 5-shot | accuracy: 74.34 -> 77.37 | Improved |
| add L_align (ablation) | GNN + L_tar + L_proxy · Places 1-shot | accuracy: 53.73 -> 54.07 | Improved |
| tune BN statistics (BN_l and BN_h) instead of no tuning (ablation) | IDP · EuroSAT 5-shot | accuracy: 88.74 -> 91.08 | Improved |
| also optimize whole backbone f_theta with alignment loss (ablation) | IDP · EuroSAT 5-shot | accuracy: 91.08 -> 88.84 | Hurt |
| also optimize whole backbone f_theta (ablation) | IDP · Places 1-shot | accuracy: 54.07 -> 53.72 | Hurt |
| tune only BN_h (high-order) vs BN_l and BN_h (ablation) | IDP · EuroSAT 1-shot | accuracy: 70.66 -> 71.60 | Improved |
| fine-tune FRN with same optimizer as IDP (ablation) | FRN · ISIC 1-shot | accuracy: 33.73 -> 33.38 | Hurt |
| fine-tune FRN with same optimizer as IDP (ablation) | FRN · CropDisease 5-shot | accuracy: 91.93 -> 90.68 | Hurt |
Closest papers by meaning
- Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.88)
- Reconstruction Target Matters in Masked Image Modeling for Cross-Domain Few-Shot Learning (AAAI 2025 · similarity 0.88)
- A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.88)
- Mind the Gap Between Prototypes and Images in Cross-domain Finetuning (NeurIPS 2024 · similarity 0.88)
- Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning (ICML 2025 · similarity 0.87)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 2025 · similarity 0.87)
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.
