Distribution-Balanced Loss for Multi-label Classification in Long-Tailed Datasets

Wu, Tong; Huang, Qingqiu; Liu, Ziwei; Wang, Yu; Lin, Dahua

doi:10.1007/978-3-030-58548-8_10

Cited by 175 publications

(133 citation statements)

References 28 publications

Supporting

Mentioning

115

Contrasting

Order By: Relevance

“…a low number of paintings), and then predict them to any given painting. Investigating different loss functions (such as focal loss [20] or distribution-balanced loss [40]) may help mitigate this issue.…”

Section: Discussionmentioning

confidence: 99%

GCNBoost: Artwork Classification by Label Propagation through a Knowledge Graph

Vaigh

García

Renoust

et al. 2021

Proceedings of the 2021 International Conference on Multimedia Retrieval

View full text Add to dashboard Cite

The rise of digitization of cultural documents offers large-scale contents, opening the road for development of AI systems in order to preserve, search, and deliver cultural heritage. To organize such cultural content also means to classify them, a task that is very familiar to modern computer science. Contextual information is often the key to structure such real world data, and we propose to use it in form of a knowledge graph. Such a knowledge graph, combined with content analysis, enhances the notion of proximity between artworks so it improves the performances in classification tasks. In this paper, we propose a novel use of a knowledge graph, that is constructed on annotated data and pseudo-labeled data. With label propagation, we boost artwork classification by training a model using a graph convolutional network, relying on the relationships between entities of the knowledge graph. Following a transductive learning framework, our experiments show that relying on a knowledge graph modeling the relations between labeled data and unlabeled data allows to achieve state-of-the-art results on multiple classification tasks on a dataset of paintings, and on a dataset of Buddha statues. Additionally, we show state-of-the-art results for the difficult case of dealing with unbalanced data, with the limitation of disregarding classes with extremely low degrees in the knowledge graph. CCS Concepts• Computing methodologies → Image representations; • Applied computing → Fine arts.

show abstract

Section: Discussionmentioning

confidence: 99%

GCNBoost: Artwork Classification by Label Propagation through a Knowledge Graph

Vaigh

García

Renoust

et al. 2021

Proceedings of the 2021 International Conference on Multimedia Retrieval

View full text Add to dashboard Cite

show abstract

“…Besides, semi-supervised learning [30] and zero-shot learning [31] are adopted for solving the problem of high cost of data labeling. The influence of loss function [32], [33] on multi-label classification is also worth studying. In [32], label-occurrence is calculated and introduced into the loss function to deal with the imbalance between positive and negative training samples.…”

Section: Related Workmentioning

confidence: 99%

“…The influence of loss function [32], [33] on multi-label classification is also worth studying. In [32], label-occurrence is calculated and introduced into the loss function to deal with the imbalance between positive and negative training samples. In [33], the binary cross-entropy function is improved by introducing scalable neighbor discriminative loss to embed a graph structure into the network.…”

Section: Related Workmentioning

confidence: 99%

Multilabel Remote Sensing Image Annotation With Multiscale Attention and Label Correlation

Huang

Zheng

Huang

2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Deep learning based multi-label image annotation is receiving increasing attention in the field of remote sensing due to the great success of deep networks in single-label remote sensing image classification. Compared with those low-level features, the features extracted by the convolutional neural network (CNN) are more informative and can alleviate the problem of semantic gap. However, the CNN model tends to ignore the smaller objects when objects of different sizes exist in an image. In addition, how to efficiently leverage the correlation among multiple labels to enhance annotation performance remains an open issue. In this paper, we propose an end-to-end deep learning framework for multi-label remote sensing image annotation. The framework is composed of a multi-scale feature fusion module, a channelspatial attention learning module and a label correlation extraction module. The multi-scale features from different layers of a CNN model are first fused and refined by using a channel-spatial attention mechanism. Then, the label correlation information is extracted from a label co-occurrence matrix, and embedded into the multi-scale attentive features to increase the discriminative ability of the resulting image features. The experiments on two benchmark data sets demonstrate the superiority of the proposed method in comparison with the state-of-the-art methods.

show abstract

“…There have been several studies that address the problem in the aspect of class imbalance. [20] alleviates data imbalance of relatively less appeared labels in multi-label samples. This suggests a re-sampling method such as [15], [16] to make class distribution uniform.…”

Section: Related Workmentioning

confidence: 99%

“…In other words, contextual bias tendency due to co-occurring becomes worse in multi-label data. In order to mitigate the co-occurring bias in multi-label data, there have been methods that reduce correlation between cooccurring instance [20] or positive-negative imbalance [35] by modifying loss function. [5] present a method of using class activation map (CAM) to reduce the CAM similarity of different objects for co-occurring samples.…”

Section: Related Workmentioning

confidence: 99%

Learning Individual Class Representation From Biased Multi-Label Data

Tserendorj

Lee

2021

IEEE Access

View full text Add to dashboard Cite

Image recognition is a popular and important research field of computer vision. Recently with the development of deep learning technology, image recognition performance has been improved significantly. However with multi-label images, recognizing individual category is a challenging task. In order to address the problem, we propose a Feature Disintegrator (FD) that decomposes co-occurred instance features of multi-label into individual categories. In experimental evaluation, proposed method achieves the gains of mean average precision (mAP) up to 18.67% and 29.05% over baseline networks in ML-MNIST and ML-CIFAR, respectively. It shows 5.76% and 6.65% higher mAP than baseline on PASCAL VOC-2007 and MS-COCO data set respectively.

show abstract

Distribution-Balanced Loss for Multi-label Classification in Long-Tailed Datasets

Cited by 175 publications

References 28 publications

GCNBoost: Artwork Classification by Label Propagation through a Knowledge Graph

GCNBoost: Artwork Classification by Label Propagation through a Knowledge Graph

Multilabel Remote Sensing Image Annotation With Multiscale Attention and Label Correlation

Learning Individual Class Representation From Biased Multi-Label Data

Contact Info

Product

Resources

About