Large Scale Adversarial Representation Learning

NeurIPS 2019 Jeff DonahueKaren Simonyan

Adversarially trained generative models (GANs) have recently achieved compelling image synthesis results. But despite early successes in using GANs for unsupervised representation learning, they have since been superseded by approaches based on self-supervision... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Self-Supervised Image Classification ImageNet BigBiGAN (RevNet-50 ×4, BN+CReLU) Top 1 Accuracy 61.3% # 31
Top 5 Accuracy 81.9% # 18
Number of Params 86M # 9
Self-Supervised Image Classification ImageNet BigBiGAN (RevNet-50 ×4) Top 1 Accuracy 60.8% # 32
Top 5 Accuracy 81.4% # 19
Self-Supervised Image Classification ImageNet BigBiGAN (ResNet-50, BN+CReLU) Top 1 Accuracy 56.6% # 36
Top 5 Accuracy 78.6% # 20
Self-Supervised Image Classification ImageNet BigBiGAN (ResNet-50) Top 1 Accuracy 55.4% # 37
Top 5 Accuracy 77.4% # 22
Semi-Supervised Image Classification ImageNet - 10% labeled data BigBiGAN (RevNet-50 ×4, BN+CReLU) Top 5 Accuracy 78.8% # 23
Semi-Supervised Image Classification ImageNet - 1% labeled data BigBiGAN (RevNet-50 ×4, BN+CReLU) Top 5 Accuracy 55.2% # 19

Methods used in the Paper