Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Variational Patch Gating for Training-Free Few-Shot Classification
Ahmed Radwan, Ahmad Abdel-Qader, Islam Osman, Mohamed Shehata
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 global CLS branch to local evidence (ablation) | VPG · CDFSL 5-way 5-shot (avg) | accuracy: 65.16 -> 67.55 | Improved |
| add local patch evidence branch to global prototype (ablation) | VPG · CDFSL 5-way 5-shot (avg) | accuracy: 66.21 -> 67.55 | Improved |
| use softplus bounded evidence instead of raw negative soft-min cost (ablation) | VPG · CDFSL 5-way 5-shot (avg) | accuracy: 66.74 -> 67.55 | Improved |
| add episodic fine-tuning of last block | VPG · CDFSL 5-way 5-shot (avg) | accuracy: 67.55 -> 71.06 | Improved |
| add episodic fine-tuning of last block | VPG · HEp 5-shot | accuracy: 69.26 -> 76.35 | Improved |
| add episodic fine-tuning of last block | VPG · BCCD_WBC 5-shot | accuracy: 68.86 -> 80.30 | Improved |
| vary foreground prior pi from 0.1 to 0.9 (ablation) | VPG · CDFSL 5-way 5-shot | accuracy: less than 1% | No clear effect |
| vary temperature epsilon from 0.01 to 0.1 (ablation) | VPG · CDFSL 5-way 5-shot | accuracy: less than 1% | No clear effect |
| reduce epsilon to 0.005 (ablation) | VPG · CDFSL EuroSAT | accuracy | Hurt |
| use parts-only (alpha=0) mixing instead of 0.4-0.6 (ablation) | VPG · CDFSL EuroSAT and CropDiseases | accuracy: loses ~2% | Hurt |
| replace H-OT with VPG (training-free) on miniImageNet 5-shot (ablation) | VPG · miniImageNet 5-way 5-shot | accuracy: 93.91 -> 96.06 | Improved |
| replace H-OT with VPG (training-free) on CUB 1-shot | VPG · CUB 5-way 1-shot | accuracy: 87.64 -> 86.02 | Hurt |
Closest papers by meaning
- Rethinking Generalization in Few-Shot Classification (NeurIPS 2022 · similarity 0.87)
- Spot and Learn: A Maximum-Entropy Patch Sampler for Few-Shot Image Classification (CVPR 2019 · similarity 0.87)
- All Patches Matter, More Patches Better: Enhance AI-Generated Image Detection via Panoptic Patch Learning (ICLR 2026 · similarity 0.87)
- Class-Aware Patch Embedding Adaptation for Few-Shot Image Classification (ICCV 2023 · similarity 0.87)
- Accelerating Convergence in Bayesian Few-Shot Classification (ICML 2024 · similarity 0.86)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · 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.
