References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: Long-tailed Recognition by Learning from Latent Categories
(Submitted on 2 Jun 2022 (v1), last revised 12 Sep 2022 (this version, v3))
Abstract: In this work, we address the challenging task of long-tailed image recognition. Previous long-tailed recognition methods commonly focus on the data augmentation or re-balancing strategy of the tail classes to give more attention to tail classes during the model training. However, due to the limited training images for tail classes, the diversity of tail class images is still restricted, which results in poor feature representations. In this work, we hypothesize that common latent features among the head and tail classes can be used to give better feature representation. Motivated by this, we introduce a Latent Categories based long-tail Recognition (LCReg) method. Specifically, we propose to learn a set of class-agnostic latent features shared among the head and tail classes. Then, we implicitly enrich the training sample diversity via applying semantic data augmentation to the latent features. Extensive experiments on five long-tailed image recognition datasets demonstrate that our proposed LCReg is able to significantly outperform previous methods and achieve state-of-the-art results.
Submission history
From: Weide Liu [view email][v1] Thu, 2 Jun 2022 12:19:51 GMT (606kb,D)
[v2] Tue, 2 Aug 2022 07:27:42 GMT (575kb,D)
[v3] Mon, 12 Sep 2022 07:05:51 GMT (577kb,D)
Link back to: arXiv, form interface, contact.