Research map / Label-efficient and robust learning / Recognition and long-tail
No Representation Rules Them All in Category Discovery
Sagar Vaze, Andrea Vedaldi, Andrew Zisserman
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 |
|---|---|---|---|
| set EMA momentum to constant 1 instead of ramped schedule (ablation) | muGCD · CUB | All ACC: 65.7 -> 1.6 | Hurt |
| set EMA momentum to 0.0 (no teacher averaging) (ablation) | muGCD · CUB | All ACC: 65.7 -> 62.7 | Hurt |
| set EMA momentum to constant 0.7 (ablation) | muGCD · CUB | All ACC: 65.7 -> 64.1 | Hurt |
| add aggressive crop and strong base augmentation to student (ablation) | muGCD · CUB | All ACC: 38.6 -> 65.7 | Improved |
| aligned instead of misaligned augmentation on color split (ablation) | GCD baseline · Clevr-4 color | All ACC: 26.1 -> 84.5 | Improved |
| aligned instead of misaligned augmentation on count split (ablation) | GCD baseline · Clevr-4 count | All ACC: 46.6 -> 65.2 | Improved |
| replace SimGCD with muGCD (teacher, staged training) | SimGCD ResNet18 · Clevr-4 average | All ACC: 80.1 -> 84.9 | Improved |
| SimGCD joint losses vs GCD baseline on texture | ResNet18 · Clevr-4 texture | All ACC: 62.4 -> 58.1 | Hurt |
| DINOv2 instead of DINO initialization | muGCD · SSB average | All ACC: 58.7 -> 72.1 | Improved |
| ResNet18 from scratch instead of DINOv2 init | SimGCD · Clevr-4 count | ACC: 51.3 -> 67.6 | Improved |
| lambda_1 between 0.1 and 0.4 vs 0 or 1 (ablation) | muGCD · Clevr-4 texture | All ACC | Mixed |
Closest papers by meaning
- Prior-Constrained Association Learning for Fine-Grained Generalized Category Discovery (AAAI 2025 · similarity 0.91)
- Hyperbolic Category Discovery (CVPR 2025 · similarity 0.90)
- Semantic-Guided Novel Category Discovery (AAAI 2024 · similarity 0.90)
- DebGCD: Debiased Learning with Distribution Guidance for Generalized Category Discovery (ICLR 2025 · similarity 0.90)
- HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts (ICLR 2025 · similarity 0.89)
- Generalized Category Discovery via Reciprocal Learning and Class-Wise Distribution Regularization (ICML 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.
