Huibin Lin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Huibin Lin has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Chun-Yang Zhang
College of Computer and Data Science, Fuzhou University, Fuzhou, China
5 shared publications - Chun Lung Philip Chen
South China University of Technology
4 shared publications - Shiping Wang
Fuzhou University, Fuzhou, China
4 shared publications - Shunxin Xiao
Fujian Key Laboratory of Pattern Recognition and Image Understanding, School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, China
3 shared publications - Wenzhong Guo
Fuzhou University
2 shared publications - Conghao Wang
College of Computing and Data Science, Nanyang Technological University, Singapore, 639798, Singapore
1 shared publication
Latest Publications
- 2 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 35 (4) , 3619-3632
Contextual Distribution Alignment via Correlation Contrasting for Domain Generalization
Domain generalization aims at learning a model with transferable knowledge from one or more source domain(s) in the presence of domain shift, enabling the model to achieve effective generalization for an unseen target domain. Most existing methods pursue domain-invariant representations of samples to address the challenges of heterogeneous distributions across domains. However, most of such methods are limited to simple data manipulation at the instance level or computing style statistics in feature space for distribution alignment. Such operations fail to effectively capture the contextual semantics across domains from both the intra and inter-views. In this paper, we propose contextual Distribution Alignment via a Contrastive Learning strategy with domain correlation, called DACL, which sufficiently exploits both intra- and inter-domain invariant representations for image domain generalization classification. Specifically, a new Fourier-based augmentation method is developed to capture high-level semantic invariant features. Second, a domain-based feature fusion module is further proposed to increase the diversity of features, which mainly extracts both intra- and inter-domain prototypes via clustering to learn cross-domain representations. Finally, we propose a contrastive learning strategy that takes domain correlation into account, which uses spatial second-order statistics as a metric to measure the relevance between multiple source domains. Extensive experiments are conducted on two domain generalization tasks over six benchmarks, demonstrating that DACL achieves state-of-the-art performance against baseline models. A series of ablation studies are performed and in-depth analyses are conducted in visualization to further verify the rationality and effectiveness of the proposed method.
- 2 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 35 (4) , 3619-3632
Contextual Distribution Alignment via Correlation Contrasting for Domain Generalization
Domain generalization aims at learning a model with transferable knowledge from one or more source domain(s) in the presence of domain shift, enabling the model to achieve effective generalization for an unseen target domain. Most existing methods pursue domain-invariant representations of samples to address the challenges of heterogeneous distributions across domains. However, most of such methods are limited to simple data manipulation at the instance level or computing style statistics in feature space for distribution alignment. Such operations fail to effectively capture the contextual semantics across domains from both the intra and inter-views. In this paper, we propose contextual Distribution Alignment via a Contrastive Learning strategy with domain correlation, called DACL, which sufficiently exploits both intra- and inter-domain invariant representations for image domain generalization classification. Specifically, a new Fourier-based augmentation method is developed to capture high-level semantic invariant features. Second, a domain-based feature fusion module is further proposed to increase the diversity of features, which mainly extracts both intra- and inter-domain prototypes via clustering to learn cross-domain representations. Finally, we propose a contrastive learning strategy that takes domain correlation into account, which uses spatial second-order statistics as a metric to measure the relevance between multiple source domains. Extensive experiments are conducted on two domain generalization tasks over six benchmarks, demonstrating that DACL achieves state-of-the-art performance against baseline models. A series of ablation studies are performed and in-depth analyses are conducted in visualization to further verify the rationality and effectiveness of the proposed method.
- 28 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 22 (15455955) , 9363-9375
Low-Overlap Point Cloud Registration via Correspondence Augmentation
Existing works have made some progress in point cloud registration, but most of them measure performance only on point cloud pairs with high overlap. In practical applications, it is often difficult to ensure that the collected point clouds overlap in large regions due to problems such as occlusion and noise. Therefore, a good low-overlap point cloud registration method is of great practical significance. However, extracting reliable correspondences from point clouds has always been a challenging task, particularly when dealing with low-overlap situation. In this paper, we propose a novel method for low-overlap point cloud registration via efficient correspondence augmentation, called AugLPCR, which not only enhances correspondences with high confidence, but also employs confidence weights to mitigate the impact of outliers. After the augmentation, the correspondences used for the transformation have a large amount of inliers, leading to improved registration performance. Extensive experiments on indoor and outdoor datasets demonstrate that the proposed AugLPCR is capable of maintaining consistent performance and achieve results comparable to or better than the state-of-the-art methods. Note to Practitioners —The motivation of this paper is to address the problem of registering two low-overlap point clouds. Mainstream algorithms for point cloud registration typically assume a sufficient overlap between point clouds. However, in practical scenarios, it is common to encounter scans with inadequate overlap. These conditions often hinder the extraction of reliable correspondences. This paper introduces an effective method for augmenting correspondences to address the problem of low inlier rates within predicted correspondences. While augmenting correspondences with high confidence, it also mitigates the influence of outliers and ambiguous points. Additionally, traditional approaches often divide superpoint regions before matching, but this can lead to the elimination of points in overlapping regions alongside outliers. To address this issue, we adjust the order of superpoint matching and region partitioning. The proposed framework can be easily applied to other correspondence-based point cloud registration models.
- 28 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 22 (15455955) , 9363-9375
Low-Overlap Point Cloud Registration via Correspondence Augmentation
Existing works have made some progress in point cloud registration, but most of them measure performance only on point cloud pairs with high overlap. In practical applications, it is often difficult to ensure that the collected point clouds overlap in large regions due to problems such as occlusion and noise. Therefore, a good low-overlap point cloud registration method is of great practical significance. However, extracting reliable correspondences from point clouds has always been a challenging task, particularly when dealing with low-overlap situation. In this paper, we propose a novel method for low-overlap point cloud registration via efficient correspondence augmentation, called AugLPCR, which not only enhances correspondences with high confidence, but also employs confidence weights to mitigate the impact of outliers. After the augmentation, the correspondences used for the transformation have a large amount of inliers, leading to improved registration performance. Extensive experiments on indoor and outdoor datasets demonstrate that the proposed AugLPCR is capable of maintaining consistent performance and achieve results comparable to or better than the state-of-the-art methods. Note to Practitioners —The motivation of this paper is to address the problem of registering two low-overlap point clouds. Mainstream algorithms for point cloud registration typically assume a sufficient overlap between point clouds. However, in practical scenarios, it is common to encounter scans with inadequate overlap. These conditions often hinder the extraction of reliable correspondences. This paper introduces an effective method for augmenting correspondences to address the problem of low inlier rates within predicted correspondences. While augmenting correspondences with high confidence, it also mitigates the influence of outliers and ambiguous points. Additionally, traditional approaches often divide superpoint regions before matching, but this can lead to the elimination of points in overlapping regions alongside outliers. To address this issue, we adjust the order of superpoint matching and region partitioning. The proposed framework can be easily applied to other correspondence-based point cloud registration models.
- 18 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (12) , 6121-6133
Unsupervised Domain Adaptation on Point Clouds via High-Order Geometric Structure Modeling
Point clouds can capture the precise geometric information of objects and scenes, which are an important source of 3D data and one of the most popular 3D geometric data structures for cognitions in many real-world applications like automatic driving and remote sensing. However, due to the influence of sensors and varieties of objects, the point clouds obtained by different devices may suffer obvious geometric changes, resulting in domain gaps that are prone to the neural networks trained in one domain failing to preserve the performance in other domains. In order to alleviate the above problem, this paper proposes an unsupervised domain adaptation framework, named HO-GSM, as the first attempt to model high-order geometric structures of point clouds. Firstly, we construct multiple self-supervised tasks to learn the invariant semantic and geometric features of the source and target domains, especially to capture the feature invariance of high-order geometric structures of point clouds. Secondly, the discriminative feature space of target domain is acquired by using contrastive learning to refine domain alignment to specific class level. Experiments on the PointDA-10 and GraspNetPC-10 collection of datasets show that the proposed HO-GSM can significantly outperform the state-of-the-art counterparts.
- 18 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (12) , 6121-6133
Unsupervised Domain Adaptation on Point Clouds via High-Order Geometric Structure Modeling
Point clouds can capture the precise geometric information of objects and scenes, which are an important source of 3D data and one of the most popular 3D geometric data structures for cognitions in many real-world applications like automatic driving and remote sensing. However, due to the influence of sensors and varieties of objects, the point clouds obtained by different devices may suffer obvious geometric changes, resulting in domain gaps that are prone to the neural networks trained in one domain failing to preserve the performance in other domains. In order to alleviate the above problem, this paper proposes an unsupervised domain adaptation framework, named HO-GSM, as the first attempt to model high-order geometric structures of point clouds. Firstly, we construct multiple self-supervised tasks to learn the invariant semantic and geometric features of the source and target domains, especially to capture the feature invariance of high-order geometric structures of point clouds. Secondly, the discriminative feature space of target domain is acquired by using contrastive learning to refine domain alignment to specific class level. Experiments on the PointDA-10 and GraspNetPC-10 collection of datasets show that the proposed HO-GSM can significantly outperform the state-of-the-art counterparts.
- 21 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 8 (5) , 3614-3627
Multi-Relation Augmentation for Graph Neural Networks
Data augmentation has been successfully utilized to refine the generalization capability and performance of learning algorithms in image and text analysis. With the rising focus on graph neural networks, an increasing number of researchers are employing various data augmentation approaches to improve graph learning techniques. Although significant improvements have been made, most of them are implemented by manipulating nodes or edges to generate modified graphs as augmented views, which might lose the information hidden in the input data. To address this issue, we propose a simple but effective data augmentation framework termed multi-relation augmentation designed for existing graph neural networks. Different from prior works, the designed model utilizes various methods to simulate multiple adjacency relationships (multi-relation) among nodes as augmented views instead of manipulating the original graph. The proposed augmentation framework can be formulated as three sub-modules, each offering distinct advantages: 1) The encoder module and projection module form a shared contrastive learning framework for both the original graph and all augmented views. Due to the shared mechanism, the proposed method can be simply applied to various graph learning models. 2) The designed task-specific module flexibly extends the proposed framework for various machine learning tasks. Experimental results on several databases show that the introduced augmentation framework improves the performance of existing graph neural networks on both semi-supervised node classification and unsupervised clustering tasks. It demonstrates that multiple relations mechanism is efficient for graph-based augmentation.
- 21 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 8 (5) , 3614-3627
Multi-Relation Augmentation for Graph Neural Networks
Data augmentation has been successfully utilized to refine the generalization capability and performance of learning algorithms in image and text analysis. With the rising focus on graph neural networks, an increasing number of researchers are employing various data augmentation approaches to improve graph learning techniques. Although significant improvements have been made, most of them are implemented by manipulating nodes or edges to generate modified graphs as augmented views, which might lose the information hidden in the input data. To address this issue, we propose a simple but effective data augmentation framework termed multi-relation augmentation designed for existing graph neural networks. Different from prior works, the designed model utilizes various methods to simulate multiple adjacency relationships (multi-relation) among nodes as augmented views instead of manipulating the original graph. The proposed augmentation framework can be formulated as three sub-modules, each offering distinct advantages: 1) The encoder module and projection module form a shared contrastive learning framework for both the original graph and all augmented views. Due to the shared mechanism, the proposed method can be simply applied to various graph learning models. 2) The designed task-specific module flexibly extends the proposed framework for various machine learning tasks. Experimental results on several databases show that the introduced augmentation framework improves the performance of existing graph neural networks on both semi-supervised node classification and unsupervised clustering tasks. It demonstrates that multiple relations mechanism is efficient for graph-based augmentation.
- 16 November 2023
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 15 (5) , 1779-1793
A new robust contrastive learning for unsupervised person re-identification
- 16 November 2023
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 15 (5) , 1779-1793
A new robust contrastive learning for unsupervised person re-identification
- 12 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 27 (9) , 4591-4600
Graph Neural Networks With Multiple Prior Knowledge for Multi-Omics Data Analysis
With the development of biotechnology, a large amount of multi-omics data have been collected for precision medicine. There exists multiple graph-based prior biological knowledge about omics data, such as gene-gene interaction networks. Recently, there has been an increasing interest in introducing graph neural networks (GNNs) into multi-omics learning. However, existing methods have not fully exploited these graphical priors since none have been able to integrate knowledge from multiple sources simultaneously. To solve this problem, we propose a multi-omics data analysis framework by incorporating multiple prior knowledge into graph neural network (MPK-GNN). To the best of our knowledge, this is the first attempt to introduce multiple prior graphs into multi-omics data analysis. Specifically, the proposed method contains four parts: (1) a feature-level learning module to aggregate information from prior graphs; (2) a projection module to maximize the agreement among prior networks by optimizing a contrastive loss; (3) a sample-level module to learn a global representation from input multi-omics features; (4) a task-specific module to flexibly extend MPK-GNN for various downstream multi-omics analysis tasks. Finally, we verify the effectiveness of the proposed multi-omics learning algorithm on the cancer molecular subtype classification task. Experimental results show that MPK-GNN outperforms other state-of-the-art algorithms, including multi-view learning methods and multi-omics integrative approaches.
- 12 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 27 (9) , 4591-4600
Graph Neural Networks With Multiple Prior Knowledge for Multi-Omics Data Analysis
With the development of biotechnology, a large amount of multi-omics data have been collected for precision medicine. There exists multiple graph-based prior biological knowledge about omics data, such as gene-gene interaction networks. Recently, there has been an increasing interest in introducing graph neural networks (GNNs) into multi-omics learning. However, existing methods have not fully exploited these graphical priors since none have been able to integrate knowledge from multiple sources simultaneously. To solve this problem, we propose a multi-omics data analysis framework by incorporating multiple prior knowledge into graph neural network (MPK-GNN). To the best of our knowledge, this is the first attempt to introduce multiple prior graphs into multi-omics data analysis. Specifically, the proposed method contains four parts: (1) a feature-level learning module to aggregate information from prior graphs; (2) a projection module to maximize the agreement among prior networks by optimizing a contrastive loss; (3) a sample-level module to learn a global representation from input multi-omics features; (4) a task-specific module to flexibly extend MPK-GNN for various downstream multi-omics analysis tasks. Finally, we verify the effectiveness of the proposed multi-omics learning algorithm on the cancer molecular subtype classification task. Experimental results show that MPK-GNN outperforms other state-of-the-art algorithms, including multi-view learning methods and multi-omics integrative approaches.
- 1 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 25 (15209210) , 8767-8779
A Probabilistic Contrastive Framework for Semi-Supervised Learning
Semi-supervised learning is a common way that investigates how to improve performance of a visual learning model, while data annotation is far from sufficient. Recent works in semi-supervised deep learning have successfully applied consistency regularization, which encourages a model to maintain consistent predictions for different perturbed versions of an image. However, most of such methods ignore the category correlation of image features, especially when exploiting strong augmentation methods for unlabeled images. To address this problem, we propose PConMatch, a model that leverages a probabilistic contrastive learning framework to separate the features of strongly-augmented versions from different classes. A semi-supervised probabilistic contrastive loss is designed, which takes both labeled and unlabeled samples into account and develops an auxiliary module to generate a probability score to measure the model prediction confidence for each sample. Specifically, PConMatch first generates a pair of weakly-augmented versions for each labeled sample, and produces a weakly-augmented version and a corresponding pair of strongly-augmented versions for each unlabeled sample. Second, a probability score module is proposed to assign pseudo-labeling confidence scores to strongly-augmented unlabeled images. Finally, the probability score of each sample is further passed to the contrastive loss, combining with consistency regularization to enable the model to learn better feature representations. Extensive experiments on four publicly available image classification benchmarks demonstrate that the proposed approach achieves state-of-the-art performance in image classification. Several rigorous ablation studies are conducted to validate the effectiveness of the method.
- 1 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 25 (15209210) , 8767-8779
A Probabilistic Contrastive Framework for Semi-Supervised Learning
Semi-supervised learning is a common way that investigates how to improve performance of a visual learning model, while data annotation is far from sufficient. Recent works in semi-supervised deep learning have successfully applied consistency regularization, which encourages a model to maintain consistent predictions for different perturbed versions of an image. However, most of such methods ignore the category correlation of image features, especially when exploiting strong augmentation methods for unlabeled images. To address this problem, we propose PConMatch, a model that leverages a probabilistic contrastive learning framework to separate the features of strongly-augmented versions from different classes. A semi-supervised probabilistic contrastive loss is designed, which takes both labeled and unlabeled samples into account and develops an auxiliary module to generate a probability score to measure the model prediction confidence for each sample. Specifically, PConMatch first generates a pair of weakly-augmented versions for each labeled sample, and produces a weakly-augmented version and a corresponding pair of strongly-augmented versions for each unlabeled sample. Second, a probability score module is proposed to assign pseudo-labeling confidence scores to strongly-augmented unlabeled images. Finally, the probability score of each sample is further passed to the contrastive loss, combining with consistency regularization to enable the model to learn better feature representations. Extensive experiments on four publicly available image classification benchmarks demonstrate that the proposed approach achieves state-of-the-art performance in image classification. Several rigorous ablation studies are conducted to validate the effectiveness of the method.
- 22 October 2021
- Published by Springer Nature
- p. 127-139
FMixAugment for Semi-supervised Learning with Consistency Regularization
- 22 October 2021
- Published by Springer Nature
- p. 127-139
FMixAugment for Semi-supervised Learning with Consistency Regularization
- Open Access
FurcaNet: An end-to-end deep gated convolutional, long short-term memory, deep neural networks for single channel speech separation
Deep gated convolutional networks have been proved to be very effective in single channel speech separation. However current state-of-the-art framework often considers training the gated convolutional networks in time-frequency (TF) domain. Such an approach will result in limited perceptual score, such as signal-to-distortion ratio (SDR) upper bound of separated utterances and also fail to exploit an end-to-end framework. In this paper we present an integrated simple and effective end-to-end approach to monaural speech separation, which consists of deep gated convolutional neural networks (GCNN) that takes the mixed utterance of two speakers and maps it to two separated utterances, where each utterance contains only one speaker's voice. In addition long short-term memory (LSTM) is employed for long term temporal modeling. For the objective, we propose to train the network by directly optimizing utterance level SDR in a permutation invariant training (PIT) style. Our experiments on the public WSJ0-2mix data corpus demonstrate that this new scheme can produce more discriminative separated utterances and leading to performance improvement on the speaker separation task.
Open AccessFurcaNet: An end-to-end deep gated convolutional, long short-term memory, deep neural networks for single channel speech separation
Deep gated convolutional networks have been proved to be very effective in single channel speech separation. However current state-of-the-art framework often considers training the gated convolutional networks in time-frequency (TF) domain. Such an approach will result in limited perceptual score, such as signal-to-distortion ratio (SDR) upper bound of separated utterances and also fail to exploit an end-to-end framework. In this paper we present an integrated simple and effective end-to-end approach to monaural speech separation, which consists of deep gated convolutional neural networks (GCNN) that takes the mixed utterance of two speakers and maps it to two separated utterances, where each utterance contains only one speaker's voice. In addition long short-term memory (LSTM) is employed for long term temporal modeling. For the objective, we propose to train the network by directly optimizing utterance level SDR in a permutation invariant training (PIT) style. Our experiments on the public WSJ0-2mix data corpus demonstrate that this new scheme can produce more discriminative separated utterances and leading to performance improvement on the speaker separation task.