Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting
With the rapid increase in data scale, real-world datasets tend to exhibit long-tailed class distributions (i.e., a few classes account for most of the data, while most classes contain only a few data points). General solutions typically exploit class rebalancing strategies involving resampling and...
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-07-01
|
| Series: | Mathematics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2227-7390/11/13/2996 |
| _version_ | 1797591228071542784 |
|---|---|
| author | Ye Yuan Jiaqi Wang Xin Xu Ruoshi Li Yongtong Zhu Lihong Wan Qingdu Li Na Liu |
| author_facet | Ye Yuan Jiaqi Wang Xin Xu Ruoshi Li Yongtong Zhu Lihong Wan Qingdu Li Na Liu |
| author_sort | Ye Yuan |
| collection | DOAJ |
| description | With the rapid increase in data scale, real-world datasets tend to exhibit long-tailed class distributions (i.e., a few classes account for most of the data, while most classes contain only a few data points). General solutions typically exploit class rebalancing strategies involving resampling and reweighting based on the sample number for each class. In this work, we explore an orthogonal direction, category splitting, which is motivated by the empirical observation that naive splitting of majority samples could alleviate the heavy imbalance between majority and minority classes. To this end, we propose a novel classwise splitting (CWS) method built upon a dynamic cluster, where classwise prototypes are updated using a moving average technique. CWS generates intra-class pseudo labels for splitting intra-class samples based on the point-to-point distance. Moreover, a group mapping module was developed to recover the ground truth of the training samples. CWS can be plugged into any existing method as a complement. Comprehensive experiments were conducted on artificially induced long-tailed image classification datasets, such as CIFAR-10-LT, CIFAR-100-LT, and OCTMNIST. Our results show that when trained with the proposed class-balanced loss, the network is able to achieve significant performance gains on long-tailed datasets. |
| first_indexed | 2024-03-11T01:34:29Z |
| format | Article |
| id | doaj.art-b5383fd980d5468489c353e21fac6b5f |
| institution | Directory Open Access Journal |
| issn | 2227-7390 |
| language | English |
| last_indexed | 2024-03-11T01:34:29Z |
| publishDate | 2023-07-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Mathematics |
| spelling | doaj.art-b5383fd980d5468489c353e21fac6b5f2023-11-18T17:04:19ZengMDPI AGMathematics2227-73902023-07-011113299610.3390/math11132996Alleviating Long-Tailed Image Classification via Dynamical Classwise SplittingYe Yuan0Jiaqi Wang1Xin Xu2Ruoshi Li3Yongtong Zhu4Lihong Wan5Qingdu Li6Na Liu7Institute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaInstitute of Machine Intelligence, University of Shanghai for Science and Technology, Shanghai 200093, ChinaWith the rapid increase in data scale, real-world datasets tend to exhibit long-tailed class distributions (i.e., a few classes account for most of the data, while most classes contain only a few data points). General solutions typically exploit class rebalancing strategies involving resampling and reweighting based on the sample number for each class. In this work, we explore an orthogonal direction, category splitting, which is motivated by the empirical observation that naive splitting of majority samples could alleviate the heavy imbalance between majority and minority classes. To this end, we propose a novel classwise splitting (CWS) method built upon a dynamic cluster, where classwise prototypes are updated using a moving average technique. CWS generates intra-class pseudo labels for splitting intra-class samples based on the point-to-point distance. Moreover, a group mapping module was developed to recover the ground truth of the training samples. CWS can be plugged into any existing method as a complement. Comprehensive experiments were conducted on artificially induced long-tailed image classification datasets, such as CIFAR-10-LT, CIFAR-100-LT, and OCTMNIST. Our results show that when trained with the proposed class-balanced loss, the network is able to achieve significant performance gains on long-tailed datasets.https://www.mdpi.com/2227-7390/11/13/2996deep learningclass-imbalance learningfeature clusteringlong-tailed classificationclasswise splitting |
| spellingShingle | Ye Yuan Jiaqi Wang Xin Xu Ruoshi Li Yongtong Zhu Lihong Wan Qingdu Li Na Liu Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting Mathematics deep learning class-imbalance learning feature clustering long-tailed classification classwise splitting |
| title | Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting |
| title_full | Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting |
| title_fullStr | Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting |
| title_full_unstemmed | Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting |
| title_short | Alleviating Long-Tailed Image Classification via Dynamical Classwise Splitting |
| title_sort | alleviating long tailed image classification via dynamical classwise splitting |
| topic | deep learning class-imbalance learning feature clustering long-tailed classification classwise splitting |
| url | https://www.mdpi.com/2227-7390/11/13/2996 |
| work_keys_str_mv | AT yeyuan alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT jiaqiwang alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT xinxu alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT ruoshili alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT yongtongzhu alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT lihongwan alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT qingduli alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting AT naliu alleviatinglongtailedimageclassificationviadynamicalclasswisesplitting |