Self-training with Noisy Student improves ImageNet classification

CVPR 2020 Qizhe XieMinh-Thang LuongEduard HovyQuoc V. Le

We present Noisy Student Training, a semi-supervised learning approach that works well even when labeled data is abundant. Noisy Student Training achieves 88.4% top-1 accuracy on ImageNet, which is 2.0% better than the state-of-the-art model that requires 3.5B weakly labeled Instagram images... (read more)

PDF Abstract

Results from the Paper


Ranked #2 on Image Classification on ImageNet (using extra training data)

     Get a GitHub badge
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT BENCHMARK
Image Classification ImageNet NoisyStudent (EfficientNet-L2) Top 1 Accuracy 88.4% # 2
Top 5 Accuracy 98.7% # 1
Number of params 480M # 4
Image Classification ImageNet NoisyStudent (EfficientNet-B7) Top 1 Accuracy 86.9% # 5
Top 5 Accuracy 98.1% # 4
Number of params 66M # 22
Image Classification ImageNet NoisyStudent (EfficientNet-B6) Top 1 Accuracy 86.4% # 7
Top 5 Accuracy 97.9% # 6
Number of params 43M # 27
Image Classification ImageNet NoisyStudent (EfficientNet-B5) Top 1 Accuracy 86.1% # 8
Top 5 Accuracy 97.8% # 7
Number of params 30M # 34
Image Classification ImageNet NoisyStudent (EfficientNet-B4) Top 1 Accuracy 85.3% # 15
Top 5 Accuracy 97.5% # 11
Number of params 19M # 45
Image Classification ImageNet NoisyStudent (EfficientNet-B3) Top 1 Accuracy 84.1% # 24
Top 5 Accuracy 96.9% # 17
Number of params 12M # 46
Image Classification ImageNet NoisyStudent (EfficientNet-B2) Top 1 Accuracy 82.4% # 39
Top 5 Accuracy 96.3% # 23
Number of params 9.2M # 48
Image Classification ImageNet NoisyStudent (EfficientNet-B1) Top 1 Accuracy 81.5% # 44
Top 5 Accuracy 95.8% # 27
Number of params 7.8M # 51
Image Classification ImageNet NoisyStudent (EfficientNet-B0) Top 1 Accuracy 78.8% # 79
Top 5 Accuracy 94.5% # 52
Number of params 5.3M # 62

Methods used in the Paper