Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Class-Aware Patch Embedding Adaptation for Few-Shot Image Classification
Fusheng Hao, Fengxiang He, Liu Liu, Fuxiang Wu, Dacheng Tao, Jun Cheng
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 class-aware adaptation of patch embeddings (lambda 0 to 2) (ablation) | CPEA · miniImageNet 1-shot | accuracy: 70.40 -> 71.97 | Improved |
| add class-aware adaptation of patch embeddings (lambda 0 to 2) (ablation) | CPEA · miniImageNet 5-shot | accuracy: 85.05 -> 87.06 | Improved |
| increase class-aware factor beyond 2 (ablation) | CPEA · miniImageNet | accuracy: 87.06 -> 87.22 (5-shot lambda 8) | No clear effect |
| add projection head (ablation) | CPEA · miniImageNet 1-shot | accuracy: 70.74 -> 71.99 | Improved |
| square the cosine similarity in the dense score matrix (ablation) | CPEA · miniImageNet 1-shot | accuracy: 71.09 -> 71.97 | Improved |
| use MIM pretraining instead of supervised DeiT (ablation) | CPEA · miniImageNet 1-shot | accuracy: 28.58 -> 71.97 | Improved |
| use MIM pretraining instead of DINO (ablation) | CPEA · miniImageNet 1-shot | accuracy: 70.65 -> 71.97 | Improved |
| use more than one class-agnostic embedding (ablation) | CPEA · miniImageNet 1-shot | accuracy: 71.97 -> 70.57 (4 embeddings) | Hurt |
| replace online-optimized FewTURE classifier with CPEA dense similarity (ablation) | ViT-S/16 backbone · miniImageNet 5-shot | accuracy: 84.51 -> 87.06 | Improved |
| replace FewTURE online optimization with CPEA | inference · miniImageNet 5-shot | inference time (ms): 168.6 -> 1.352 | Improved |
| increase input resolution from 224 to 448 (ablation) | CPEA · miniImageNet 5-shot | accuracy: 87.06 -> 88.00 | Improved |
| replace cosine similarity with scaled Euclidean metrics (ablation) | CPEA · miniImageNet 5-shot | accuracy: 87.06 -> 85.78 (best Euclidean variant) | Hurt |
Closest papers by meaning
- Rethinking Generalization in Few-Shot Classification (NeurIPS 2022 · similarity 0.88)
- Latent Embeddings for Zero-Shot Classification (CVPR 2016 · similarity 0.88)
- Variational Patch Gating for Training-Free Few-Shot Classification (ECCV 2026 · similarity 0.87)
- Bayesian Test-Time Adaptation for Vision-Language Models (CVPR 2025 · similarity 0.85)
- Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement (ICLR 2025 · similarity 0.85)
- Follow-Up Differential Descriptions: Language Models Resolve Ambiguities for Image Classification (ICLR 2024 · similarity 0.85)
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.
