TC-Net: Detecting Noisy Labels Via Transform Consistency
- 1 October 2021
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 24 (15209210) , 4328-4341
Abstract
It is crucial to distinguish mislabeled samples for dealing with noisy labels. Previous methods such as “Co-teaching” and “JoCoR” introduce two different networks to select clean samples out of the noisy ones and only use these clean samples to train the deep models. Different from these methods which require to train two networks simultaneously, we propose a simple and effective method to identify clean samples only using one single network. We discover that the clean samples prefer to reach consistent predictions for the original images and the transformed images while noisy samples usually suffer from inconsistent predictions. Motivated by this observation, we propose a noisy label detection approach, named Transform Consistency Network (TC-Net), which constrains the transform consistency ( i.e., category consistency and visual attention consistency) between the original images and the transformed images for network training. Then we can select small-loss samples to update the parameters of the network. Furthermore, in order to mitigate the negative influence of noisy labels, we design a classification loss by using the off-line hard labels and on-line soft labels to provide more reliable supervisions for training a robust model. We conduct comprehensive experiments on CIFAR-10, CIFAR-100 and Clothing1M datasets. Compared with the clean sample selection baselines, we achieve the state-of-the-art performance. Especially, in most cases, our proposed method outperforms the baselines by a large margin.Keywords
Funding Information
- Fundamental Research Funds for the Central Universities (2019JBZ104)
- Major Research and Development Plan of China State Railway Group Company Ltd. (K2020G024)
This publication has 17 references indexed in Scilit:
- Visual Attention Consistency Under Image Transforms for Multi-Label Image Classification2019
- MetaCleaner: Learning to Hallucinate Clean Representations for Noisy-Labeled Visual Recognition2019
- Learning to Learn From Noisy Labeled Data2019
- CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images2018
- An Efficient and Provable Approach for Mixture Proportion Estimation Using Linear Independence AssumptionIEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2018
- Joint Optimization Framework for Learning with Noisy Labels2018
- Learning Deep Features for Discriminative Localization2016
- Fast R-CNN2015
- Classification with Noisy Labels by Importance ReweightingIEEE Transactions on Pattern Analysis and Machine Intelligence, 2015
- Learning from massive noisy labeled data for image classification2015