Solving Long-Tailed Recognition with Deep Realistic Taxonomic Classifier

Tie-jun, WU; Morgado, Pedro; Wang, Pei; Ho, Chih-Hui; Vasconcelos, Nuno

doi:10.1007/978-3-030-58598-3_11

Cited by 24 publications

(16 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Long-tailed object detector with classification equilibrium (LOCE) [33] proposed to use the mean classification prediction score (i.e., running prediction probability) to monitor model training on different classes, and guide memory-augmented feature sampling for enhancing tail-class performance. [67] CVPR 2016 Focal loss [68] ICCV 2017 Range loss [21] ICCV 2017 CRL [69] ICCV 2017 MetaModelNet [70] NeurIPS 2017 DSTL [71] CVPR 2018 CB [16] CVPR 2019 Bayesian estimate [72] CVPR 2019 FTL [73] CVPR 2019 Unequal-training [74] CVPR 2019 OLTR [15] CVPR 2019 DCL [75] ICCV 2019 Meta-Weight-Net [76] NeurIPS 2019 LDAM [18] NeurIPS 2019 Decoupling [32] ICLR 2020 LST [77] CVPR 2020 BBN [48] CVPR 2020 BAGS [78] CVPR 2020 Domain adaptation [28] CVPR 2020 Equalization loss (ESQL) [19] CVPR 2020 DBM [22] CVPR 2020 M2m [79] CVPR 2020 LEAP [80] CVPR 2020 IEM [81] CVPR 2020 SimCal [34] ECCV 2020 PRS [82] ECCV 2020 Distribution-balanced loss [37] ECCV 2020 OFA [83] ECCV 2020 LFME [84] ECCV 2020 Deep-RTC [85] ECCV 2020 Balanced Meta-Softmax [86] NeurIPS 2020 UNO-IC [87] NeurIPS 2020 De-confound-TDE [88] NeurIPS 2020 SSP [89] NeurIPS 2020 Logit adjustment [14] ICLR 2021 RIDE [17] ICLR 2021 KCL [13] ICLR 2021 LTML [90] CVPR 2021 Equalization loss v2…”

Section: Re-samplingmentioning

confidence: 99%

See 1 more Smart Citation

Deep Long-Tailed Learning: A Survey

Zhang¹,

Kang²,

Hooi³

et al. 2021

Preprint

View full text Add to dashboard Cite

Deep long-tailed learning, one of the most challenging problems in visual recognition, aims to train well-performing deep models from a large number of images that follow a long-tailed class distribution. In the last decade, deep learning has emerged as a powerful recognition model for learning high-quality image representations and has led to remarkable breakthroughs in generic visual recognition. However, long-tailed class imbalance, a common problem in practical visual recognition tasks, often limits the practicality of deep network based recognition models in real-world applications, since they can be easily biased towards dominant classes and perform poorly on tail classes. To address this problem, a large number of studies have been conducted in recent years, making promising progress in the field of deep long-tailed learning. Considering the rapid evolution of this field, this paper aims to provide a comprehensive survey on recent advances in deep long-tailed learning. To be specific, we group existing deep long-tailed learning studies into three main categories (i.e., class re-balancing, information augmentation and module improvement), and review these methods following this taxonomy in detail. Afterward, we empirically analyze several state-of-the-art methods by evaluating to what extent they address the issue of class imbalance via a newly proposed evaluation metric, i.e., relative accuracy. We conclude the survey by highlighting important applications of deep long-tailed learning and identifying several promising directions for future research.

show abstract

Section: Re-samplingmentioning

confidence: 99%

“…Realistic taxonomic classifier (RTC) [85] proposed to address class imbalance with hierarchical classification. Specifically, RTC maps images into a class taxonomic tree structure, where the hierarchy is defined by a set of classification nodes and node relations.…”

Section: Classifier Designmentioning

confidence: 99%

Deep Long-Tailed Learning: A Survey

Zhang¹,

Kang²,

Hooi³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

Section: Related Workmentioning

confidence: 99%

“…Hierarchical classification typically employs a label hierarchy in the form of a tree [8,6,5,22,1,18] or directed acyclic graph [7,12] that explicitly injects prior knowledge of the label relationships into the model. The relationship between the labels can be either 'IS-A' or not depending on the associated classification problem [1,8,7,6,5,22,18,3,2]. One could either define the label hierarchy manually with domain knowledge [1,12,18] or derive the hierarchy automatically [8,6,5,7,22] from a well established semantic lexical database, such as WordNet [16].…”

Section: Related Workmentioning

confidence: 99%

“…The relationship between the labels can be either 'IS-A' or not depending on the associated classification problem [1,8,7,6,5,22,18,3,2]. One could either define the label hierarchy manually with domain knowledge [1,12,18] or derive the hierarchy automatically [8,6,5,7,22] from a well established semantic lexical database, such as WordNet [16].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Bottom-up Hierarchical Classification Using Confusion-based Logit Compression

Liang,

Davis,

Ilin

2021

Preprint

View full text Add to dashboard Cite

In this work, we propose a method to efficiently compute label posteriors of a base flat classifier in the presence of few validation examples within a bottom-up hierarchical inference framework. A stand-alone validation set (not used to train the base classifier) is preferred for posterior estimation to avoid overfitting the base classifier, however a small validation set limits the number of features one can effectively use. We propose a simple, yet robust, logit vector compression approach based on generalized logits [6] and label confusions for the task of label posterior estimation within the context of hierarchical classification. Extensive comparative experiments with other compression techniques are provided across multiple sized validation sets, and a comparison with related hierarchical classification approaches is also conducted. The proposed approach mitigates the problem of not having enough validation examples for reliable posterior estimation while maintaining strong hierarchical classification performance.

show abstract