We introduce Bootstrap Your Own Latent (BYOL), a new approach to self-supervised image representation learning. BYOL relies on two neural networks, referred to as online and target networks, that interact and learn from each other... From an augmented view of an image, we train the online network to predict the target network representation of the same image under a different augmented view. At the same time, we update the target network with a slow-moving average of the online network. While state-of-the art methods rely on negative pairs, BYOL achieves a new state of the art without them. BYOL reaches $74.3\%$ top-1 classification accuracy on ImageNet using a linear evaluation with a ResNet-50 architecture and $79.6\%$ with a larger ResNet. We show that BYOL performs on par or better than the current state of the art on both transfer and semi-supervised benchmarks. Our implementation and pretrained models are given on GitHub. read more

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Self-Supervised Image Classification ImageNet BYOL (ResNet-200 x2) Top 1 Accuracy 79.6% # 10
Top 5 Accuracy 94.8% # 3
Number of Params 250M # 16
Self-Supervised Image Classification ImageNet BYOL (ResNet-50) Top 1 Accuracy 74.3% # 45
Top 5 Accuracy 91.6% # 16
Self-Supervised Image Classification ImageNet BYOL (ResNet-50 x2) Top 1 Accuracy 77.4% # 23
Top 5 Accuracy 93.6% # 7
Number of Params 94M # 19
Self-Supervised Image Classification ImageNet BYOL (ResNet-50 x4) Top 1 Accuracy 78.6% # 15
Top 5 Accuracy 94.2% # 5
Number of Params 375M # 9
Semi-Supervised Image Classification ImageNet - 1% labeled data BYOL (ResNet-50 x2) Top 5 Accuracy 84.1% # 13
Top 1 Accuracy 62.2% # 13
Semi-Supervised Image Classification ImageNet - 1% labeled data BYOL (ResNet-50) Top 5 Accuracy 78.4% # 20
Top 1 Accuracy 53.2% # 21
Semi-Supervised Image Classification ImageNet - 1% labeled data BYOL (ResNet-200 x2) Top 5 Accuracy 89.5% # 5
Top 1 Accuracy 71.2% # 5
Semi-Supervised Image Classification ImageNet - 1% labeled data BYOL (ResNet-50 x4) Top 5 Accuracy 87.9% # 7
Top 1 Accuracy 69.1% # 6
Image Classification Places205 BYOL Top 1 Accuracy 54.0 # 5
Person Re-Identification SYSU-30k BYOL (self-supervised) Rank-1 12.7 # 6
Self-Supervised Person Re-Identification SYSU-30k BYOL Rank-1 12.7 # 2

Methods