Research map / Vision-language and multimodal / Cross-modal learning
SpectralGCD: Spectral Concept Selection and Cross-modal Representation Learning for Generalized Category Discovery
Lorenzo Caselli, Marco Mistretta, Simone Magistri, Andrew D. Bagdanov
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 forward and reverse distillation from frozen teacher (ablation) | SpectralGCD · Stanford Cars (All) | clustering accuracy: 77.4 -> 89.1 | Improved |
| add reverse distillation to forward distillation (ablation) | SpectralGCD · Stanford Cars (All) | clustering accuracy: 86.0 -> 89.1 | Improved |
| add forward distillation to reverse distillation (ablation) | SpectralGCD · CUB (All) | clustering accuracy: 79.3 -> 79.2 | No clear effect |
| train classifier on cross-modal representation instead of image features (ablation) | SpectralGCD · CUB (All) | clustering accuracy: 76.3 -> 79.2 | Improved |
| train classifier on cross-modal representation instead of image features (ablation) | SpectralGCD · CIFAR-100 (Old) | clustering accuracy (Old): 88.7 -> 87.2 | Hurt |
| apply Spectral Filtering to the concept dictionary (ablation) | SpectralGCD · Stanford Cars / CIFAR-100 | clustering accuracy (All) | Improved |
| use stronger teacher (DFN-5B H/14 vs ViT-B/16) (ablation) | SpectralGCD · CIFAR-100 (All) | clustering accuracy: 78.5 -> 88.8 | Improved |
| use Tags dictionary instead of OpenImagesV7 (ablation) | SpectralGCD · Stanford Cars (All) | clustering accuracy: 85.8 -> 89.1 | Improved |
| use ViT-H/14 student instead of ViT-B/16 (ablation) | SpectralGCD · CUB (All) | clustering accuracy: 79.2 -> 80.0 | Improved |
| compute spectral filtering from Old samples only (ablation) | SpectralGCD · CUB (All) | clustering accuracy: 79.2 -> 77.8 | Hurt |
| use SpectralGCD instead of TextGCD (Tags+Attributes) | TextGCD · Stanford Cars (All) | clustering accuracy: 86.9 -> 89.1 | Improved |
| use SpectralGCD instead of TextGCD | TextGCD · CIFAR-10/100 (All) | clustering accuracy: 98.2 -> 98.5; 85.7 -> 86.1 | No clear effect |
Closest papers by meaning
- DebGCD: Debiased Learning with Distribution Guidance for Generalized Category Discovery (ICLR 2025 · similarity 0.89)
- HiLo: A Learning Framework for Generalized Category Discovery Robust to Domain Shifts (ICLR 2025 · similarity 0.88)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.88)
- SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning (ICLR 2024 · similarity 0.88)
- Generalized Category Discovery under Domain Shift: A Frequency Domain Perspective (NeurIPS 2025 · similarity 0.87)
- Generalized Category Discovery via Reciprocal Learning and Class-Wise Distribution Regularization (ICML 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.
