Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Reconstruction Guided Meta-Learning for Few Shot Open Set Recognition
Sayak Nag, Dripta S. Raychaudhuri, Sujoy Paul, Amit K. Roy–Chowdhury
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 |
|---|---|---|---|
| use VAE instead of autoencoder for reconstruction (ablation) | ReFOCS · GTSRB->TT100K 5-way 5-shot | AUROC: 77.08 -> 85.25 | Improved |
| use exemplar-weighted prototype instead of simple centroid (ablation) | ReFOCS · GTSRB->TT100K 5-way 5-shot | accuracy: 81.79 -> 83.36 | Improved |
| use exemplar-weighted prototype instead of simple centroid (ablation) | ReFOCS · GTSRB->TT100K 5-way 5-shot | AUROC: 84.93 -> 85.25 | Improved |
| add embedding modulation (ablation) | ReFOCS · Belga->Flickr32 5-way 1-shot | AUROC: 56.69 -> 72.98 | Improved |
| add exemplar reconstruction errors as detector input (ablation) | ReFOCS · Belga->Flickr32 5-way 1-shot | AUROC: 50.01 -> 72.98 | Improved |
| add embedding as detector input (ablation) | ReFOCS · GTSRB->TT100K 5-way 5-shot | AUROC: 74.33 -> 85.25 | Improved |
| add classification score as detector input (ablation) | ReFOCS · GTSRB->TT100K 5-way 5-shot | AUROC: 82.85 -> 85.25 | Improved |
| use cosine instead of Euclidean distance (ablation) | ReFOCS · Belga->Flickr32 5-way 1-shot | AUROC: 50.61 -> 72.9 | Improved |
| use cosine instead of Euclidean distance (ablation) | ReFOCS · miniImageNet 5-way 5-shot (Resnet18) | accuracy: 79.09 vs 79.06 | No clear effect |
| use canonical instead of estimated exemplars | ReFOCS · GTSRB->TT100K 5-way 5-shot | AUROC: 79.48 -> 85.25 | Improved |
| use ReFOCS instead of PEELER | ReFOCS · GTSRB->TT100K (openness 29.3%) | F1: 40.28 -> 52.31 | Improved |
| use ReFOCS instead of PEELER | ReFOCS · Belga->Flickr32 5-way 1-shot (openness 0%) | F1: 48.91 -> 53.65 | Improved |
Closest papers by meaning
- Few-Shot Open-Set Recognition Using Meta-Learning (CVPR 2020 · similarity 0.92)
- Few-Shot Open-Set Recognition by Transformation Consistency (CVPR 2021 · similarity 0.90)
- Meta Evidential Transformer for Few-Shot Open-Set Recognition (ICML 2024 · similarity 0.89)
- Recent Advances in Open Set Recognition: A Survey (TPAMI 2020 · similarity 0.89)
- Open-Set Likelihood Maximization for Few-Shot Learning (CVPR 2023 · similarity 0.89)
- 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.
