Research map / Label-efficient and robust learning / Recognition and long-tail
TAR: Token-Aware Refinement for Fine-grained Generalized Category Discovery
Xingyu Yang, Yu Zhang, Siya Mi, Xiu-Shen Wei
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 TAR to GET | GET (CLIP ViT-B/16) · CUB (All) | clustering accuracy: 77.0 -> 78.4 | Improved |
| add TAR to MOS | MOS (DINO ViT-B/16) · CUB (All) | clustering accuracy: 69.6 -> 70.6 | Improved |
| add TAR to GET on Herbarium 19 | GET · Herbarium 19 (All) | clustering accuracy: 49.7 -> 49.8 | No clear effect |
| add TAR to GET on ImageNet-1k | GET · ImageNet-1k (All) | clustering accuracy: 62.4 -> 63.1 | Improved |
| add TAR to GET on FGVC-Aircraft novel classes | GET · FGVC-Aircraft (Novel) | clustering accuracy: +0.0 | No clear effect |
| add TRM alone (non-parametric, no learnable parameters) (ablation) | GET · CUB (All) | clustering accuracy: 75.5 -> 75.5 | No clear effect |
| add GRM with TRM but without CATS weighting (ablation) | GET · CUB (All) | clustering accuracy: 75.5 -> 74.0 | Hurt |
| add CATS without GRM (ablation) | GET · CUB (All) | clustering accuracy: 75.5 -> 74.3 | Hurt |
| add consistency regularization loss (ablation) | TAR · CUB (All) | clustering accuracy: 75.9 -> 78.4 | Improved |
| add KL divergence loss (ablation) | TAR · CUB (All) | clustering accuracy: 75.3 -> 78.4 | Improved |
| add Token Reweighting Module (ablation) | TAR · CUB (All) | clustering accuracy: 76.7 -> 78.4 | Improved |
| add Gate Refinement Module (ablation) | TAR · CUB (All) | clustering accuracy: 76.1 -> 78.4 | Improved |
Closest papers by meaning
- When Domain Generalization meets Generalized Category Discovery: An Adaptive Task-Arithmetic Driven Approach (CVPR 2025 · similarity 0.87)
- Transfer and Alignment Network for Generalized Category Discovery (AAAI 2024 · similarity 0.87)
- Consistent Supervised-Unsupervised Alignment for Generalized Category Discovery (NeurIPS 2025 · similarity 0.87)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.87)
- HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts (ICLR 2025 · similarity 0.87)
- Less Attention is More: Prompt Transformer for Generalized Category Discovery (CVPR 2025 · similarity 0.87)
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.
