Self-Supervised Learning by Estimating Twin Class Distributions

14 Oct 2021  ·  Feng Wang, Tao Kong, Rufeng Zhang, Huaping Liu, Hang Li ·

We present TWIST, a simple and theoretically explainable self-supervised representation learning method by classifying large-scale unlabeled datasets in an end-to-end way. We employ a siamese network terminated by a softmax operation to produce twin class distributions of two augmented images. Without supervision, we enforce the class distributions of different augmentations to be consistent. However, simply minimizing the divergence between augmentations will cause collapsed solutions, i.e., outputting the same class probability distribution for all images. In this case, no information about the input image is left. To solve this problem, we propose to maximize the mutual information between the input and the class predictions. Specifically, we minimize the entropy of the distribution for each sample to make the class prediction for each sample assertive and maximize the entropy of the mean distribution to make the predictions of different samples diverse. In this way, TWIST can naturally avoid the collapsed solutions without specific designs such as asymmetric network, stop-gradient operation, or momentum encoder. As a result, TWIST outperforms state-of-the-art methods on a wide range of tasks. Especially, TWIST performs surprisingly well on semi-supervised learning, achieving 61.2% top-1 accuracy with 1% ImageNet labels using a ResNet-50 as backbone, surpassing previous best results by an absolute improvement of 6.2%. Codes and pre-trained models are given on: https://github.com/bytedance/TWIST

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Fine-Grained Image Classification Caltech-101 TWIST (ResNet-50 ) Top-1 Error Rate 6.5% # 5
Accuracy 93.5% # 3
Image Classification DTD TWIST (ResNet-50) Accuracy 76.6 # 7
Image Classification Food-101 TWIST (ResNet-50) Accuracy (%) 89.3 # 3
Semi-Supervised Image Classification ImageNet - 10% labeled data TWIST (ResNet-50 x2) Top 5 Accuracy 92.8% # 6
Top 1 Accuracy 75.3% # 18
Semi-Supervised Image Classification ImageNet - 1% labeled data TWIST (ResNet-50 x2) Top 5 Accuracy 88.2% # 7
Top 1 Accuracy 67.2% # 16
Image Classification Oxford-IIIT Pet Dataset TWIST (ResNet-50) Accuracy 94.5 # 1
Fine-Grained Image Classification SUN397 TWIST (ResNet-50) Accuracy 67.4 # 4

Methods