Unsupervised Feature Learning via Non-Parametric Instance Discrimination

CVPR 2018 Zhirong WuYuanjun XiongStella X. YuDahua Lin

Neural net classifiers trained on data with annotated class labels can also capture apparent visual similarity among categories without being directed to do so. We study whether this observation can be extended beyond the conventional domain of supervised learning: Can we learn a good feature representation that captures apparent similarity among instances, instead of classes, by merely asking the feature to be discriminative of individual instances?.. (read more)

PDF Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK BENCHMARK
Self-Supervised Image Classification ImageNet InstDisc (ResNet-50) Top 1 Accuracy 54.0% # 40
Top 1 Accuracy (kNN) 46.5% # 5
Semi-Supervised Image Classification ImageNet - 10% labeled data InstDisc (ResNet-50) Top 5 Accuracy 77.4% # 25
Semi-Supervised Image Classification ImageNet - 10% labeled data Instance Discrimination Top 5 Accuracy 77.40% # 25
Semi-Supervised Image Classification ImageNet - 1% labeled data Instance Discrimination (ResNet-50) Top 5 Accuracy 39.20% # 27

Methods used in the Paper