Open llien30 opened 4 years ago
Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Han Zhang, Colin Raffel
Google Research
2020.1
New data augmentation method
Train a model with few labeled data\ → Input weakly-augumentated unlabeled image and makes Pseudo-label of the image\ → Input strongly-augumentated version of the same image and train a model with a standard cross-entropy loss
Only retain an artificial label if the model assigns a high probability to one of the possible classes
2020/6/12
INFO
Author
Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Han Zhang, Colin Raffel
Affiliation
Google Research
Conference or Year
2020.1
Link
Abstract
New data augmentation method
Proposed Method
Train a model with few labeled data\ → Input weakly-augumentated unlabeled image and makes Pseudo-label of the image\ → Input strongly-augumentated version of the same image and train a model with a standard cross-entropy loss
Only retain an artificial label if the model assigns a high probability to one of the possible classes
Evaluation
Contribution
Discussion, Future Work
Comment
Date
2020/6/12