Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Variational Feature Disentangling for Fine-Grained Few-Shot Classification
Jingyi Xu, Hieu Le, Mingzhen Huang, ShahRukh Athar, Dimitris Samaras
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 augmented features from the instance-conditioned posterior with 1NN (ablation) | VFD · CUB 5-way 1NN | accuracy: 74.95 -> 75.46 | Improved |
| add augmented features from Delta-encoder with 1NN | Delta-encoder · CUB 5-way 1NN | accuracy: 69.37 -> 67.31 | Hurt |
| add Delta-encoder augmentation at 20-way with 1NN | Delta-encoder · CUB 20-way 1NN | accuracy: 48.10 -> 38.84 | Hurt |
| add our augmentation at 10-way with 1NN (ablation) | VFD · CUB 10-way 1NN | accuracy: 62.05 -> 63.17 | Improved |
| sample variance features from instance-conditioned posterior instead of N(0,I) (ablation) | VFD · CUB and NAB 5-way 1-shot | accuracy | Improved |
| use isotropic Gaussian prior instead of Gaussian mixture prior (ablation) | VFD · CUB 5-way 1-shot | accuracy: 75.16 -> 79.12 | Improved |
| use variational inference instead of covariance matrix estimate (ablation) | VFD · CUB 5-way 1-shot | accuracy: 75.28 -> 79.12 | Improved |
| add feature disentanglement of intra-class variance (ablation) | VFD · CUB 5-way 1-shot | accuracy: 73.40 -> 79.12 | Improved |
| add feature disentanglement of intra-class variance (ablation) | VFD · NAB 5-way 1-shot | accuracy: 81.83 -> 88.62 | Improved |
| use VFD instead of MetaOptNet | VFD · CUB 5-way 5-shot (ResNet12) | accuracy: 87.09 -> 91.48 | Improved |
| use VFD instead of Delta-encoder as augmentation | VFD · Stanford Dogs 5-way 5-shot (ResNet12) | accuracy: 78.60 -> 88.00 | Improved |
| use VFD instead of MattML | VFD · CUB 5-way 1-shot (Conv4) | accuracy: 66.29 -> 68.42 | 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.
- Delta-encoder — Delta-encoder: an effective sample synthesis method for few-shot object recognition (NeurIPS 2018) · better on 6 of 6 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
- A Closer Look at Few-shot Classification (ICLR 2019 · similarity 0.88)
- Bi-Directional Ensemble Feature Reconstruction Network for Few-Shot Fine-Grained Classification (TPAMI 2024 · similarity 0.88)
- Bi-directional Feature Reconstruction Network for Fine-Grained Few-Shot Image Classification (AAAI 2023 · similarity 0.88)
- Hierarchical Disentanglement of Discriminative Latent Features for Zero-Shot Learning (CVPR 2019 · similarity 0.87)
- Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning (CVPR 2024 · similarity 0.86)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (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.
