Research map / Label-efficient and robust learning / Self-supervised and contrastive
SCAN: Learning to Classify Images without Labels
Wouter Van Gansbeke, Simon Vandenhende, Stamatios Georgoulis, Marc Proesmans, Luc Van Gool
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 |
|---|---|---|---|
| replace K-means with SCAN-loss clustering (SimCLR augmentation) (ablation) | Pretext features · CIFAR10 | accuracy: 65.9 -> 78.7 | Improved |
| use RandAugment instead of SimCLR augmentations in SCAN-loss (ablation) | SCAN-loss · CIFAR10 | accuracy: 78.7 -> 81.8 | Improved |
| add self-labeling after clustering (ablation) | SCAN · CIFAR10 | accuracy: 81.8 -> 87.6 | Improved |
| use strong augmentation (RA) rather than SimCLR augmentation in self-labeling (ablation) | SCAN self-labeling · CIFAR10 | accuracy: 10.0 -> 87.4 | Improved |
| use instance discrimination pretext instead of RotNet (ablation) | SCAN · CIFAR10 | accuracy: 74.3 -> 87.6 | Improved |
| use instance discrimination (SimCLR) pretext instead of RotNet with K-means (ablation) | K-means · CIFAR10 | accuracy: 27.1 -> 65.9 | Improved |
| use nearest neighbors K=5 instead of K=0 (ablation) | SCAN · CIFAR10 | accuracy: 56.3 -> 79.3 | Improved |
| use nearest neighbors K=5 instead of K=0 (ablation) | SCAN · CIFAR100-20 | accuracy: 24.6 -> 41.1 | Improved |
| use nearest neighbors K=5 instead of K=0 (ablation) | SCAN · STL10 | accuracy: 47.70 -> 69.8 | Improved |
| overcluster to 2x the number of classes | SCAN · CIFAR100-20 | accuracy: 45.9 -> 55.1 | Improved |
| overcluster to 2x the number of classes | SCAN · CIFAR10 | accuracy: 87.6 -> 86.2 | No clear effect |
| overcluster to 2x the number of classes | SCAN · STL10 | accuracy: 76.7 -> 76.8 | No clear effect |
Closest papers by meaning
- Invariant Information Clustering for Unsupervised Image Classification and Segmentation (ICCV 2019 · similarity 0.92)
- Unsupervised Few-Shot Image Classification by Learning Features into Clustering Space (ECCV 2022 · similarity 0.91)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.90)
- Semantic-Enhanced Image Clustering (AAAI 2023 · similarity 0.89)
- Automatically Discovering and Learning New Visual Categories with Ranking Statistics (ICLR 2020 · similarity 0.89)
- Language-Assisted Image Clustering Guided by Discriminative Relational Signals and Adaptive Semantic Centers (NeurIPS 2026 · 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.
