Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Milking CowMask for Semi-Supervised Image Classification

Geoff French, Avital Oliver, Tim Salimans

TL;DR

This work tackles the labeling bottleneck in image classification by leveraging consistency regularization in a semi-supervised setting. It introduces CowMask, a flexible mask-based perturbation, and two modes—CowOut (erasure) and CowMix (mixing)—within a Mean Teacher framework. CowMix achieves state-of-the-art results on ImageNet with 10% labeled data (Top-1 26.06%, Top-5 8.76%), and the approach shows competitive performance on CIFAR-10/100 and SVHN, demonstrating broad applicability. The method is simple, end-to-end, and open-sourced, offering a practical path to improved semi-supervised learning without complex training schedules.

Abstract

Consistency regularization is a technique for semi-supervised learning that underlies a number of strong results for classification with few labeled data. It works by encouraging a learned model to be robust to perturbations on unlabeled data. Here, we present a novel mask-based augmentation method called CowMask. Using it to provide perturbations for semi-supervised consistency regularization, we achieve a state-of-the-art result on ImageNet with 10% labeled data, with a top-5 error of 8.76% and top-1 error of 26.06%. Moreover, we do so with a method that is much simpler than many alternatives. We further investigate the behavior of CowMask for semi-supervised learning by running many smaller scale experiments on the SVHN, CIFAR-10 and CIFAR-100 data sets, where we achieve results competitive with the state of the art, indicating that CowMask is widely applicable. We open source our code at https://github.com/google-research/google-research/tree/master/milking_cowmask

Milking CowMask for Semi-Supervised Image Classification

TL;DR

This work tackles the labeling bottleneck in image classification by leveraging consistency regularization in a semi-supervised setting. It introduces CowMask, a flexible mask-based perturbation, and two modes—CowOut (erasure) and CowMix (mixing)—within a Mean Teacher framework. CowMix achieves state-of-the-art results on ImageNet with 10% labeled data (Top-1 26.06%, Top-5 8.76%), and the approach shows competitive performance on CIFAR-10/100 and SVHN, demonstrating broad applicability. The method is simple, end-to-end, and open-sourced, offering a practical path to improved semi-supervised learning without complex training schedules.

Abstract

Consistency regularization is a technique for semi-supervised learning that underlies a number of strong results for classification with few labeled data. It works by encouraging a learned model to be robust to perturbations on unlabeled data. Here, we present a novel mask-based augmentation method called CowMask. Using it to provide perturbations for semi-supervised consistency regularization, we achieve a state-of-the-art result on ImageNet with 10% labeled data, with a top-5 error of 8.76% and top-1 error of 26.06%. Moreover, we do so with a method that is much simpler than many alternatives. We further investigate the behavior of CowMask for semi-supervised learning by running many smaller scale experiments on the SVHN, CIFAR-10 and CIFAR-100 data sets, where we achieve results competitive with the state of the art, indicating that CowMask is widely applicable. We open source our code at https://github.com/google-research/google-research/tree/master/milking_cowmask

Paper Structure

This paper contains 19 sections, 1 equation, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Semi-supervised classification
  4. Mixing regularization
  5. Rich augmentation
  6. Mask-based regularization
  7. CowMask
  8. Semi-Supervised Learning Method
  9. Mask-based augmentation by erasure
  10. Mask-based mixing
  11. Experiments and results
  12. ImageNet 2012
  13. Setup
  14. Results
  15. Small image experiments
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Illustration of the unsupervised mask based erasure consistency loss component of semi-supervised image classification. Blue arrows carry image or mask content and grey arrows carry probability vectors. Note that confidence thresholding is not illustrated here.
  • Figure 2: Illustration of the unsupervised masked based mixing loss component of semi-supervised image classification. Blue arrows carry image or mask content, grey arrows carry probability vectors and yellow carry scalars. Please note that confidence thresholding is not illustrated here.