Research map / Label-efficient and robust learning / Domain adaptation and generalization
Reclaiming Lost Text Layers for Source-Free Cross-Domain Few-Shot Learning
Zhenyu Zhang, Guangyao Chen, 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 VtT to CLIP-LoRA-Vision | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 55.97 -> 58.23 | Improved |
| add VtT to Maple | Maple · CDFSL 5-way 1-shot avg | accuracy: 53.07 -> 55.15 | Improved |
| add VtT to SigLIP2-LoRA and PE-Core-LoRA backbones | SigLIP2-LoRA / PE-Core-LoRA · CDFSL 5-way 5-shot avg | accuracy: 65.96 -> 67.75; 68.64 -> 70.05 | Improved |
| emphasize the lost text layer instead of removing it (ablation) | CLIP fine-tuning · CDFSL 5-way 1-shot avg | accuracy: 56.3 -> 56.7 | Improved |
| choose the wrong layer to remove (ablation) | Removal strategy · CDFSL 5-way 1-shot avg | accuracy: 56.3 -> 50.7 | Hurt |
| add TIA module (ablation) | CLIP-LoRA-Vision · CDFSL 5-way 1-shot avg | accuracy: 55.9 -> 56.9 | Improved |
| add V-T Fusion on top of TIA (ablation) | VtT · CDFSL 5-way 1-shot avg | accuracy: 56.9 -> 57.6 | Improved |
| add DGSO on top of TIA + V-T Fusion (ablation) | VtT · CDFSL 5-way 1-shot avg | accuracy: 57.6 -> 58.2 | Improved |
| use deep-to-shallow instead of shallow-to-deep scanning (ablation) | V-T Fusion · CDFSL 5-way 1-shot avg | accuracy: 57.7 -> 58.2 | Improved |
| use bidirectional scanning instead of deep-to-shallow (ablation) | V-T Fusion · CDFSL 5-way 1-shot avg | accuracy: 58.2 -> 58.2 | No clear effect |
| fuse both visual and text features rather than one modality (ablation) | V-T Fusion · CDFSL 5-way 1-shot avg | accuracy: 57.1 / 57.5 -> 58.2 | Improved |
| use SSM fusion instead of mean averaging (ablation) | V-T Fusion · CDFSL 5-way 1-shot avg | accuracy: 57.0 -> 58.2 | Improved |
Closest papers by meaning
- A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.89)
- Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.88)
- Attention Temperature Matters in ViT-Based Cross-Domain Few-Shot Learning (NeurIPS 2024 · similarity 0.88)
- Reconstruction Target Matters in Masked Image Modeling for Cross-Domain Few-Shot Learning (AAAI 2025 · similarity 0.88)
- Random Registers for Cross-Domain Few-Shot Learning (ICML 2025 · similarity 0.88)
- Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning (ICML 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.
