Research map / Label-efficient and robust learning / Recognition and long-tail
Adaptive Gaussian Expansion for On-the-fly Category Discovery
Chunming Li, Shidong Wang, Haofeng Zhang
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 open-set recognition (OSR) component (ablation) | AGE · CUB-200 | Accuracy (All, %): 38.8 -> 46.3 | Improved |
| add soft covariance (SC) (ablation) | AGE · CUB-200 | Accuracy (All, %): 27.0 -> 46.3 | Improved |
| add soft covariance (SC) without LW (ablation) | AGE · CUB-200 | Accuracy (All, %): 26.2 -> 45.0 | Improved |
| add Ledoit-Wolf covariance estimation (ablation) | AGE · CUB-200 | Accuracy (All, %): 45.0 -> 46.3 | Improved |
| add soft covariance and Ledoit-Wolf (ablation) | AGE · ImageNet-100 | Accuracy (All, %): 35.2 -> 48.2 | Improved |
| use random projection instead of PCA (ablation) | AGE · ImageNet-100 | Accuracy (All, %): 48.2 -> 52.2 | Improved |
| use random projection instead of PCA (ablation) | AGE · CUB-200 | Accuracy (All, %): 46.3 -> 40.1 | Hurt |
| increase stabilization term epsilon from 1e-5 to 1e-2 (ablation) | AGE · CUB-200 | Accuracy (All, %): 46.3 -> 30.9 | Hurt |
| use epsilon 1e-7 instead of 1e-5 (ablation) | AGE · CUB-200 | Accuracy (All, %): 46.3 -> 45.9 | No clear effect |
| keep or mask old-class samples or inject noise into Gaussian estimation (ablation) | AGE · CUB-200 new-class accuracy | Accuracy (New classes, %): 43.5 -> 43.9 | No clear effect |
| use AGE instead of PHE | PHE · Herbarium19 OCD | Accuracy (All, %): 22.5 -> 30.9 | Improved |
| use AGE instead of PHE | PHE · iNaturalist Fungi | Accuracy (All, %): 31.4 -> 40.0 | Improved |
Closest papers by meaning
- TALON: Test-time Adaptive Learning for On-the-Fly Category Discovery (CVPR 2026 · similarity 0.90)
- On-the-Fly Category Discovery (CVPR 2023 · similarity 0.89)
- Prototypical Hash Encoding for On-the-Fly Fine-Grained Category Discovery (NeurIPS 2024 · similarity 0.88)
- Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (NeurIPS 2024 · similarity 0.86)
- Active Generalized Category Discovery (CVPR 2024 · similarity 0.86)
- Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective (ICML 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.
