Filters








3,370 Hits in 3.5 sec

Decoupling Representation and Classifier for Long-Tailed Recognition [article]

Bingyi Kang, Saining Xie, Marcus Rohrbach, Zhicheng Yan, Albert Gordo, Jiashi Feng, Yannis Kalantidis
2020 arXiv   pre-print
In this work, we decouple the learning procedure into representation learning and classification, and systematically explore how different balancing strategies affect them for long-tailed recognition.  ...  it is also possible to achieve strong long-tailed recognition ability by adjusting only the classifier.  ...  LEARNING REPRESENTATIONS FOR LONG-TAILED RECOGNITION For long-tailed recognition, the training set follows a long-tailed distribution over the classes.  ... 
arXiv:1910.09217v2 fatcat:dnqprca7ebdf3iymafr63hsxje

Deep Long-Tailed Learning: A Survey [article]

Yifan Zhang, Bingyi Kang, Bryan Hooi, Shuicheng Yan, Jiashi Feng
2021 arXiv   pre-print
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.  ...  We conclude the survey by highlighting important applications of deep long-tailed learning and identifying several promising directions for future research.  ...  We begin with Decoupling [32] , which empirically evaluated various sampling strategies for representation learning on long-tailed recognition.  ... 
arXiv:2110.04596v1 fatcat:lpvt2x6cv5crxm2qxdctjrlkqq

Margin Calibration for Long-Tailed Visual Recognition [article]

Yidong Wang, Bowen Zhang, Wenxin Hou, Zhen Wu, Jindong Wang, Takahiro Shinozaki
2022 arXiv   pre-print
The long-tailed class distribution in visual recognition tasks poses great challenges for neural networks on how to handle the biased predictions between head and tail classes, i.e., the model tends to  ...  We hope this simple method will motivate people to rethink the biased margins and biased logits in long-tailed visual recognition.  ...  For other datasets, we directly use 1.2 for γ. Conclusions This paper studied the long-tailed visual recognition problem.  ... 
arXiv:2112.07225v4 fatcat:6du2wzkqjjbqhnxozxxieblqta

Relieving the Incompatibility of Network Representation and Classification for Long-Tailed Data Distribution

Hao Hu, Mengya Gao, Mingsheng Wu, Syed Hassan Ahmed
2021 Computational Intelligence and Neuroscience  
In this paper, we use knowledge distillation to solve the long-tailed data distribution problem and fully optimize the network representation and classifier simultaneously.  ...  Recent work focuses on improving the network representation ability to overcome the long-tailed problem, while it always ignores adapting the network classifier to a long-tailed case, which will cause  ...  Methods For long-tailed recognition, the training dataset follows an imbalance distribution over classes.  ... 
doi:10.1155/2021/6702625 pmid:34987568 pmcid:PMC8723848 fatcat:r2s5dw3vrnff5hi5pc5ym6sdii

The Devil is the Classifier: Investigating Long Tail Relation Classification with Decoupling Analysis [article]

Haiyang Yu, Ningyu Zhang, Shumin Deng, Zonggang Yuan, Yantao Jia, Huajun Chen
2020 arXiv   pre-print
for all classes; moreover, it is possible to achieve better long-tailed classification ability at low cost by only adjusting the classifier.  ...  ., data re-sampling and loss re-weighting, but all these methods adhere to the schema of entangling learning of the representation and classifier.  ...  ther research on promising directions, including 1) exploiting a more effective classifier to boost long-tailed discriminability, 2) distinguishing taskspecific representation, and classifier automatically  ... 
arXiv:2009.07022v1 fatcat:xdzmlheahvemva6nycx2gnuqfq

Investigate the Essence of Long-Tailed Recognition from a Unified Perspective [article]

Lei Liu, Li Liu
2021 arXiv   pre-print
Specifically, we demonstrate that long-tailed recognition suffers from both sample number and category similarity.  ...  representation methods (e.g., self-supervised learning) for similarity reduction, the classifier bias can be further alleviated with greatly improved performance.  ...  Indeed, for the long-tailed recognition, decouple training (or two-stage training) [14] is known to perform recent SOTA results for long-tailed recognition.  ... 
arXiv:2107.03758v1 fatcat:ze3k3azahrgmtna6xraf5jhghq

Self Supervision to Distillation for Long-Tailed Visual Recognition [article]

Tianhao Li, Limin Wang, Gangshan Wu
2021 arXiv   pre-print
We conduct extensive experiments and our method achieves the state-of-the-art results on three long-tailed recognition benchmarks: ImageNet-LT, CIFAR100-LT and iNaturalist 2018.  ...  First, we introduce a self-distillation framework for long-tailed recognition, which can mine the label relation automatically.  ...  To overcome these issues, the recent work [20, 48] decouples the tasks of representation learning and classifier training.  ... 
arXiv:2109.04075v1 fatcat:5gi4s55pcfaixbcrwfvss5erym

Improving Tail-Class Representation with Centroid Contrastive Learning [article]

Anthony Meng Huat Tiong, Junnan Li, Guosheng Lin, Boyang Li, Caiming Xiong, Steven C.H. Hoi
2021 arXiv   pre-print
Recent developments have shown good long-tailed model can be learnt by decoupling the training into representation learning and classifier balancing.  ...  This distribution poses difficulty in learning good representations for tail classes.  ...  However, long-tailed recognition remains as one of the major challenges.  ... 
arXiv:2110.10048v1 fatcat:p3iroctzs5hflfu3zbob6mnwyy

A Weight Moving Average Based Alternate Decoupled Learning Algorithm for Long-Tailed Language Identification

Hui Wang, Lin Liu, Yan Song, Lei Fang, Ian McLoughlin, Li-Rong Dai
2021 Conference of the International Speech Communication Association  
This raises the challenge of long-tailed LID. In this paper, we propose an effective weight moving average (WMA) based alternate decoupled learning algorithm, termed WADCL, for long-tailed LID.  ...  The system is divided into two components, a frontend feature extractor and a backend classifier.  ...  (b) Proposed WMA based alternate decoupled learning framework for long-tailed LID task.  ... 
doi:10.21437/interspeech.2021-776 dblp:conf/interspeech/WangLSF0021 fatcat:6r6uxlsiyjeb5o3hvv3qaouaiq

Calibrating Class Activation Maps for Long-Tailed Visual Recognition [article]

Chi Zhang, Guosheng Lin, Lvlong Lai, Henghui Ding, Qingyao Wu
2021 arXiv   pre-print
Furthermore, we investigate the use of normalized classifiers for representation learning in long-tailed problems.  ...  Real-world visual recognition problems often exhibit long-tailed distributions, where the amount of data for learning in different categories shows significant imbalance.  ...  (S), RG22/19 (S) and RG95/20.  ... 
arXiv:2108.12757v1 fatcat:7j3gchou4zejxidnijpqwl2suy

Balanced Contrastive Learning for Long-Tailed Visual Recognition [article]

Jianggang, Zhu and Zheng, Wang and Jingjing, Chen and Yi-Ping Phoebe, Chen and Yu-Gang, Jiang
2022 arXiv   pre-print
To correct the optimization behavior of SCL and further improve the performance of long-tailed visual recognition, we propose a novel loss for balanced contrastive learning (BCL).  ...  However, through our theoretical analysis, we find that for long-tailed data, it fails to form a regular simplex which is an ideal geometric configuration for representation learning.  ...  Acknowledgements This work was supported in part by NSFC project (# 62072116), Shanghai Municipal Commission of Economy and Informatization Project (2020-GYHLW-01009), and in part by Shanghai Pujiang Program  ... 
arXiv:2207.09052v1 fatcat:x3ppcaxyr5flpn76wslwiljmb4

Recognition and Processing of NATOM [article]

YiPeng Deng, YinHui Luo
2021 arXiv   pre-print
Using Glove word vector methods to represent the data for using a custom mapping vocabulary. 2.Decoupling features and classifiers: In order to improve the ability of the text classification model to recognize  ...  The weights of the feature learning stage and the classifier learning stage adopt different strategies to overcome the influence of the head data and tail data of the imbalanced data set on the classification  ...  The seventh is the decoupling representation and classifier [9] (decoupling representation and classifier). The text classification process is divided into two stages.  ... 
arXiv:2105.03314v1 fatcat:5643jt74arghdcae3dodms3ady

Improving Calibration for Long-Tailed Recognition [article]

Zhisheng Zhong, Jiequan Cui, Shu Liu, Jiaya Jia
2021 arXiv   pre-print
Our proposed methods set new records on multiple popular long-tailed recognition benchmark datasets, including CIFAR-10-LT, CIFAR-100-LT, ImageNet-LT, Places-LT, and iNaturalist 2018.  ...  Recently, two-stage methods decouple representation learning and classifier learning to improve performance. But there is still the vital issue of miscalibration.  ...  for long-tailed recognition.  ... 
arXiv:2104.00466v1 fatcat:npygsrv7uva43g5enoukqlo7sq

You Only Need End-to-End Training for Long-Tailed Recognition [article]

Zhiwei Zhang
2021 arXiv   pre-print
Experimental results on the long-tailed classification benchmarks, CIFAR-LT and ImageNet-LT, demonstrate the effectiveness of our method.  ...  The generalization gap on the long-tailed data sets is largely owing to most categories only occupying a few training samples.  ...  The proposed end-to-end training framework for long-tailed recognition.  ... 
arXiv:2112.05958v3 fatcat:ekaw5tjaezetvlmmzpira6dv5u

CoSSL: Co-Learning of Representation and Classifier for Imbalanced Semi-Supervised Learning [article]

Yue Fan and Dengxin Dai and Anna Kukleva and Bernt Schiele
2022 arXiv   pre-print
In this paper, we propose a novel co-learning framework (CoSSL) with decoupled representation learning and classifier learning for imbalanced SSL.  ...  To handle the data imbalance, we devise Tail-class Feature Enhancement (TFE) for classifier learning.  ...  This evaluation protocol can be used for long-tailed recognition as well.  ... 
arXiv:2112.04564v3 fatcat:cyfqgnld75hyrpcj5q2siptcca
« Previous Showing results 1 — 15 out of 3,370 results