Milking CowMask for Semi-Supervised Image Classification

By Geoff French, Avital Oliver, Tim Salimans. 2020

60 citations recorded by the source corpus.

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 this https URL

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Sanja Fidler
2. Antonio Torralba
3. Karsten Kreis
4. Daiqing Li
5. Luc Van Gool
6. Lennart Svensson
7. Wilhelm Tranheden
8. Viktor Olsson

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Sanja Fidler
3. Karsten Kreis
4. Antonio Torralba
5. Ming-Hsuan Yang
6. Jitendra Malik
7. Dacheng Tao
8. Trevor Darrell

In Radiology, Nuclear Medicine and Imaging

1. Antonio Torralba
2. Lennart Svensson
3. Wilhelm Tranheden
4. Viktor Olsson
5. Karsten Kreis
6. Sanja Fidler
7. Juliano Pinto
8. Luc Van Gool

In Representation & Self-Supervised Learning

1. Antonio Torralba
2. Sanja Fidler
3. Lennart Svensson
4. Wilhelm Tranheden
5. Viktor Olsson
6. Daiqing Li
7. Karsten Kreis
8. Juliano Pinto

Citations in the corpus

• DACS: Domain Adaptation via Cross-domain Mixed Sampling (2020, methodology): “When replacing ClassMix [29] with other mixing techniques such as CutMix [49], or CowMix [12], performance drops slightly but the class conflation issue remains solved independent of mixing strategy, as can be seen in Table 3.”
• Semantic Segmentation with Generative Models: Semi-Supervised Learning and Strong Out-of-Domain Generalization (2021, methodology): “The literature on SSL is vast and some classical SSL techniques include pseudolabeling [50, 2, 8, 83], consistency regularization [80, 49, 87, 23, 83], and various data augmentation techniques [7, 6, 91] (also see Sec.”