Research map / Architectures and efficiency / Attention and transformers
A Hidden Stumbling Block in Generalized Category Discovery: Distracted Attention
Qiyu Xu, Zhanxuan Hu, Yu Duan, Ercheng Pei, Yonghang Tai
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 AF (attention focusing) to SimGCD | SimGCD · Stanford Cars | ACC (All): 53.8 -> 67.0 | Improved |
| add AF to SimGCD | SimGCD · CUB | ACC (All): 60.3 -> 69.0 | Improved |
| add AF to SimGCD | SimGCD · FGVC-Aircraft | ACC (All): 54.2 -> 59.4 | Improved |
| add AF to SimGCD | SimGCD · CIFAR100 | ACC (New): 77.8 -> 76.5 | Hurt |
| add AF to SimGCD | SimGCD · Herbarium-19 | ACC (All): 44.0 -> 45.5 | Improved |
| add AF to SimGCD | SimGCD · CIFAR10 | ACC (All): 97.1 -> 97.8 | Improved |
| add AF to other GCD methods | SelEx · CUB | ACC (All): 73.4 -> 79.2 | Improved |
| add AF to other GCD methods | CMS · Stanford Cars | ACC (All): 53.1 -> 61.8 | Improved |
| learn queries on all training data instead of labeled data only (ablation) | SimGCD+AF · CUB | ACC (All): 69.0 -> 67.4 | Hurt |
| apply TAP to both views (multi-view) instead of single view (ablation) | SimGCD+AF · Stanford Cars | ACC (All): 67.0 -> 63.2 | Hurt |
| use fixed-k pruning (k=128) instead of adaptive TAP (ablation) | SimGCD+AF · Stanford Cars | ACC (All): 67.0 -> 62.4 | Hurt |
| use [CLS] token attention scores instead of TIME importance (ablation) | SimGCD+AF · CUB | ACC (All): 69.0 -> 63.9 | Hurt |
Closest papers by meaning
- Less Attention is More: Prompt Transformer for Generalized Category Discovery (CVPR 2025 · similarity 0.88)
- Transfer and Alignment Network for Generalized Category Discovery (AAAI 2024 · similarity 0.87)
- Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · similarity 0.87)
- SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning (ICLR 2024 · similarity 0.87)
- Active Generalized Category Discovery (CVPR 2024 · similarity 0.87)
- Generalized Category Discovery under Domain Shift: A Frequency Domain Perspective (NeurIPS 2025 · 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.
