Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
PatchMix Augmentation to Identify Causal Features in Few-Shot Learning
Chengming Xu, Chen Liu, Xinwei Sun, Siqian Yang, Yabiao Wang, Chengjie Wang, Yanwei Fu
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 Mixup/CutMix with PatchMix augmentation (ablation) | FSL baseline · miniImageNet 5-way | accuracy (%): CutMix 65.91 -> PatchMix 68.34 (1-shot), 79.10 -> 83.16 (5-shot) | Improved |
| add Manifold Mixup to baseline (ablation) | FSL baseline · miniImageNet 5-way 5-shot | accuracy (%): 77.95 vs baseline decrease of 1.39% | Hurt |
| add CutMix to CAN (ablation) | CAN · miniImageNet 5-way 5-shot | accuracy (%): 81.42 -> 79.53 | Hurt |
| add PatchMix to CAN (ablation) | CAN · miniImageNet 5-way | accuracy (%): 65.05 -> 67.77 (1-shot), 81.42 -> 82.54 (5-shot) | Improved |
| add correlation-guided reconstruction (gumbel-softmax) after PatchMix (ablation) | PatchMix · miniImageNet 5-way | accuracy (%): 68.34 -> 69.38 (1-shot), 83.16 -> 84.14 (5-shot) | Improved |
| add vanilla reconstruction (ablation) | PatchMix · miniImageNet 5-way | accuracy (%): 68.34 -> 68.59 | No clear effect |
| use softmax instead of gumbel-softmax in CGR (ablation) | CGR · miniImageNet 5-way | accuracy (%): 69.38 -> 68.81 | Hurt |
| add global hardness to PatchMix distillation (ablation) | hardness-aware PatchMix · miniImageNet 5-way | accuracy (%): 68.61 -> 69.38 (1-shot), 83.46 -> 84.14 (5-shot) | Improved |
| use global instead of local hardness (ablation) | hardness-aware PatchMix · miniImageNet 5-way 1-shot | accuracy (%): 69.07 -> 69.38 | Improved |
| use all mixed images instead of mix plus original (ablation) | PatchMix · miniImageNet 5-way 1-shot | accuracy (%): 68.06 -> 68.61 | Improved |
| use grid size 11x11 instead of 6x6 (ablation) | PatchMix · miniImageNet 5-way 1-shot | accuracy (%): 67.95 -> 68.61 | Improved |
| replace DeepCluster with PatchMix in unsupervised FSL (ablation) | CACTUs-ProtoNets · miniImageNet unsupervised 1-shot | accuracy (%): 40.69 -> 41.73 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepEMD — DeepEMD: Few-Shot Image Classification With Differentiable Earth Mover's Distance and Structured Classifiers (CVPR 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Interventional Few-Shot Learning (NeurIPS 2020 · similarity 0.90)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.86)
- Text Augmented Correlation Transformer For Few-shot Classification & Segmentation (CVPR 2025 · similarity 0.86)
- ZeroDiff: Solidified Visual-semantic Correlation in Zero-Shot Learning (ICLR 2025 · similarity 0.85)
- Causal Inference over Visual-Semantic-Aligned Graph for Image Classification (AAAI 2025 · similarity 0.85)
- Visual Representation Learning through Causal Intervention for Controllable Image Editing (CVPR 2025 · 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.
