Unsupervised Learning of Visual Features by Contrasting Cluster Assignments

arxiv:2006.09882

Original source · Paper JSON

Self-supervised learning and vision · 2020-06-17

Cites (0) · Cited by (10)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3

Contents

• Unsupervised Learning of Visual Features by Contrasting Cluster Assignments
• Abstract
• 1 Introduction
• 2 Related Work
• Instance and contrastive learning.
• Clustering for deep representation learning.
• Handcrafted pretext tasks.
• 3 Method
• 3.1 Online clustering
• Swapped prediction problem.
• Computing codes online.
• Working with small batches.
• 3.2 Multi-crop: Augmenting views with smaller images
• 4 Main Results
• 4.1 Evaluating the unsupervised features on ImageNet
• 4.2 Transferring unsupervised features to downstream tasks
• 4.3 Training with small batches
• 5 Ablation Study
• 5.1 Clustering-based self-supervised learning
• Improving prior clustering-based approaches.
• Comparing clustering with contrastive instance learning.
• Advantage of SwAV compared to DeepCluster-v2.
• 5.2 Applying the multi-crop strategy to different methods
• 5.3 Impact of longer training
• 5.4 Unsupervised pretraining on a large uncurated dataset
• 6 Discussion
• Acknowledgement.