Research map / Label-efficient and robust learning / Self-supervised and contrastive
Unsupervised Learning of Visual Features by Contrasting Cluster Assignments
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| learn prototypes instead of fixing them (ablation) | SwAV · ImageNet linear | top-1: 73.1 -> 73.9 | Helped |
| soft codes instead of hard assignments (ablation) | SwAV · ImageNet linear | top-1: 73.3 -> 73.9 | Helped |
| increase number of prototypes beyond 3000 (ablation) | SwAV · ImageNet linear | top-1: 73.9 -> 73.8 | No clear effect |
| more Sinkhorn iterations (3 -> 30) (ablation) | SwAV · ImageNet linear | top-1: 73.9 -> 73.7 | No clear effect |
| add multi-crop augmentation (ablation) | SimCLR, DeepCluster, SeLa, SwAV · ImageNet linear | top-1: +2-4% | Helped |
| add multi-crop augmentation (ablation) | supervised model · ImageNet | top-1 | No clear effect |
| clustering-based objective instead of contrastive instance (ablation) | SimCLR setup · ImageNet linear | top-1: +2% / +3.5% with multi-crop | Helped |
| longer training (100 -> 800 epochs) (ablation) | SwAV · ImageNet linear | top-1: 72.1 -> 75.3 | Helped |
| SwAV pretraining instead of supervised | ResNet-50 · COCO Mask R-CNN | box AP: 39.7 -> 41.6 | Helped |
Closest papers by meaning
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.88)
- Poly-View Contrastive Learning (ICLR 2024 · similarity 0.87)
- Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning (CVPR 2022 · similarity 0.87)
- What Should Not Be Contrastive in Contrastive Learning (ICLR 2021 · similarity 0.87)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.86)
- Contrastive Tuning: A Little Help to Make Masked Autoencoders Forget (AAAI 2024 · similarity 0.85)
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.