Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
CrossTransformers: spatially-aware few-shot transfer
Carl Doersch, Ankush Gupta, Andrew Zisserman
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 |
|---|---|---|---|
| replace global prototypes with spatially aligned local-feature comparisons | Prototypical Nets · Meta-Dataset Train-on-ILSVRC | average rank: ProtoNets (Our implementation): 11.10; CTX: 3.85 | Improved |
| add 50% SimCLR episodes (ablation) | CrossTransformers · Meta-Dataset | accuracy: OmniGlot: 76.52 -> 80.83; Aircraft: 79.65 -> 82.05; CU-Birds: 84.06 -> 82.01 | Mixed |
| add BOHB-style augmentation to CTX+SimCLR episodes (ablation) | CrossTransformers · Meta-Dataset | accuracy: OmniGlot: 80.83 -> 82.21; Quick Draw: 68.84 -> 72.68; Aircraft: 82.05 -> 79.49 | Mixed |
| add test-time logistic regression on globally pooled features (ablation) | CTX+SimCLR Eps+Aug · DTD | accuracy: 75.57 -> 80.29 | Improved |
| add test-time logistic regression on globally pooled features (ablation) | CTX+SimCLR Eps+Aug · Aircraft | accuracy: 79.49 -> 77.41 | Hurt |
| add SimCLR instance-discrimination episodes rather than only SimCLR-style augmentation (ablation) | Prototypical Nets · Meta-Dataset | accuracy | Improved |
| apply SimCLR-style augmentation without instance discrimination (ablation) | Prototypical Nets · Meta-Dataset | accuracy | Hurt |
| increase Prototypical Net capacity and image resolution (ablation) | Prototypical Nets · Meta-Dataset | accuracy | Mixed |
| compute Batch Norm statistics from the test-time support set instead of using EMA (ablation) | Prototypical Nets with SimCLR episodes · Meta-Dataset | accuracy | Improved |
| increase feature-map resolution from 7×7 to 14×14 (ablation) | CrossTransformers · Meta-Dataset | accuracy: Aircraft: 70.90 -> 79.65; CU-Birds: 80.29 -> 84.06; OmniGlot: 74.11 -> 76.52 | Improved |
| increase feature-map resolution from 7×7 to 14×14 (ablation) | CrossTransformers · OmniGlot | accuracy: 87.26 -> 82.21 | Hurt |
Closest papers by meaning
- CrossFormer: A Versatile Vision Transformer Hinging on Cross-scale Attention (ICLR 2022 · similarity 0.89)
- CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification (ICCV 2021 · similarity 0.88)
- A Theoretical Analysis of Self-Supervised Learning for Vision Transformers (ICLR 2025 · similarity 0.87)
- An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels (ICLR 2025 · similarity 0.86)
- Revisiting Continuity of Image Tokens for Cross-domain Few-shot Learning (ICML 2025 · similarity 0.86)
- LaVin-DiT: Large Vision Diffusion Transformer (CVPR 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.
