Research map / Continual learning / Class-incremental
Virtual Category-Guided Continual Generalized Category Discovery
Jiahui Xiong, Qiuxia Lai, Hongsong Wang
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 virtual category learning (VCL) (ablation) | ENCL-less baseline · CIFAR-100 | avg accuracy (All): 69.00 -> 70.04 | Improved |
| add virtual category learning (VCL) (ablation) | baseline · Tiny-ImageNet | avg accuracy (All): 63.22 -> 63.13 | No clear effect |
| add expanded neighborhood contrastive learning (ENCL) (ablation) | baseline · CIFAR-100 | avg accuracy (All): 69.00 -> 69.57 | Improved |
| add expanded neighborhood contrastive learning (ENCL) (ablation) | baseline · Tiny-ImageNet | avg accuracy (All): 63.22 -> 64.22 | Improved |
| add ENCL on top of VCL (ablation) | VCL · CIFAR-100 | avg accuracy (All): 70.04 -> 71.27 | Improved |
| add VCL on top of ENCL (ablation) | ENCL · Tiny-ImageNet | avg accuracy (All): 64.22 -> 64.40 | Improved |
| add VCL on top of ENCL (ablation) | ENCL · CIFAR-100 | New accuracy: 51.10 -> 56.61 | Improved |
| ambiguity threshold 0.15 instead of 0.07 (ablation) | VCL · CIFAR-100 | avg accuracy: 68.63 -> 71.27 | Improved |
| ambiguity threshold 0.15 instead of 0.3 (ablation) | VCL · CIFAR-100 | avg accuracy: 68.42 -> 71.27 | Improved |
| neighborhood size 3 instead of 4 (ablation) | ENCL · CIFAR-100 | avg accuracy: 69.09 -> 71.27 | Improved |
| neighborhood size 3 instead of 2 (ablation) | ENCL · CIFAR-100 | avg accuracy: 70.52 -> 71.27 | Improved |
| add ENCL (training cost) (ablation) | model without ENCL · CIFAR-100 | training time s/epoch: 338 -> 384 | Hurt |
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.
- MetaGCD — MetaGCD: Learning to Continually Learn in Generalized Category Discovery (ICCV 2023) · better on 3 of 3 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
- Beyond the Static World: Continual Category Discovery under Visual Drift (CVPR 2026 · similarity 0.93)
- MetaGCD: Learning to Continually Learn in Generalized Category Discovery (ICCV 2023 · similarity 0.93)
- Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (NeurIPS 2024 · similarity 0.91)
- Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective (ICML 2025 · similarity 0.91)
- Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · similarity 0.90)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.89)
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.
