JXJunsheng Xiao

Junsheng Xiao

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Junsheng Xiao has a total of 30 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • SwinBTC: Transfer Learning to Brain Tumor Classification for Healthcare Electronics Using Augmented MR Images

    by Honghao Gao, Yaping Wan, Hongxia Xu, Lingchao Chen, Junsheng Xiao, Qionghuizi Ran

    Brain tumors require AI-assisted, precise treatments. Methods based on healthcare electronics, such as magnetic resonance imaging (MRI), computed tomography (CT), and gastrointestinal endoscopy, are widely used in hospitals. Maximizing the functionality of these electronic devices is crucial for improving tumor lesion detection and is a challenge in AI-assisted medical applications. Precise classification is vital for effectively planning brain tumor treatments, and accurate classification results offer crucial insights that enable physicians to devise optimal treatment strategies. Therefore, this paper proposes a novel brain tumor classification method named SwinBTC, which is based on the healthcare Internet of Things (HIoT) and integrates a pretrained and fine-tuned Swin transformer model to improve the performance of healthcare electronics. First, a transfer learning (TL)-based brain tumor classification model called SwinBTC is proposed. The general features in the images are used to improve the classification ability of the brain tumor MRI results, further accelerate the model training speed, and avoid overfitting problems. Second, clinical magnetic resonance images obtained from HIoT nodes are used to enrich the dataset, and online and offline data augmentation techniques are used to expand the utilized dataset and increase data diversity, improving the generalizability and reliability of the developed model. Finally, the performance of the proposed approach is evaluated on the CE-MRI and TT-MRI datasets via various classification metrics, and the experimental results reveal that our method outperforms other baselines.


  • Toward Effective 3D Object Detection via Multimodal Fusion to Automatic Driving for Industrial Cyber-Physical Systems

    by Honghao Gao, Yan Sun, Junsheng Xiao, Danqing Fang, Yueshen Xu, Wei Wei

    AI-empowered automatic driving has experienced rapid development in industrial cyber-physical systems (CPSs), especially in safety vehicles and driverless technologies. 3D object detection is an important task for perceiving the surrounding environment and supporting decision-making when vehicles are on the road, and is also a focus in CPSs. Light detection and ranging (LiDAR)-based detection methods usually lack semantic information, resulting in high uncertainty with incorrect outputs. Thus, handling complex road scenes is difficult. Some data fusion-based methods have been developed to solve these issues. However, the spatiotemporal data misalignment between different sensors is prone to losing information during data fusion. This paper proposes exploiting multimodal information to learn more high-level features to address these issues thus reducing the uncertainty of 3D object detection. First, the VxMLA (voxel and multilevel attention) framework is employed to improve point cloud identification and modeling during 3D object detection. Second, the MF-CAMRL (modal fusion-based channel attention and multidimensional regression loss) model is proposed with two subnetworks. Our model encompasses two strategies, i.e., a multimodal fusion and a deep learning model based on CAMRL. One focuses on the semantic complementarity and geometric proximity for decisionlevel fusion. The other focuses on weighted ensemble bounding boxes to fully utilize the highlevel decision information derived from both modalities and reduce the information loss incurred during modal fusion. Finally, sufficient experiments are performed on the KITTI dataset and presented. The results show that our method is superior to baseline methods.


  • Open Access

    Few-Shot Object Detection with Memory Contrastive Proposal Based on Semantic Priors

    by Linlin Xiao, Huahu Xu, Junsheng Xiao, Yuzhe Huang

    Few-shot object detection (FSOD) aims to detect objects belonging to novel classes with few training samples. With the small number of novel class samples, the visual information extracted is insufficient to accurately represent the object itself, presenting significant intra-class variance and confusion between classes of similar samples, resulting in large errors in the detection results of the novel class samples. We propose a few-shot object detection framework to achieve effective classification and detection by embedding semantic information and contrastive learning. Firstly, we introduced a semantic fusion (SF) module, which projects semantic spatial information into visual space for interaction, to compensate for the lack of visual information and further enhance the representation of feature information. To further improve the classification performance, we embed the memory contrastive proposal (MCP) module to adjust the distribution of the feature space by calculating the contrastive loss between the class-centered features of previous samples and the current input features to obtain a more discriminative embedding space for better intra-class aggregation and inter-class separation for subsequent classification and detection. Extensive experiments on the PASCAL VOC and MS-COCO datasets show that the performance of our proposed method is effectively improved. Our proposed method improves nAP50 over the baseline model by 4.5% and 3.5%.

  • FS3DCIoT: A Few-Shot Incremental Learning Network for Skin Disease Differential Diagnosis in the Consumer IoT

    by Junsheng Xiao, Jirui Li, Honghao Gao

    The computer-aided diagnosis (CAD) method based on few-shot learning (FSL) effectively reduces the dependence on labelled medical images. However, the catastrophic forgetting defect of neural networks seriously decreases the ability of CAD methods to meet real diagnosis needs. Moreover, the low classification accuracy and limited categories make FSL classification difficult. A few-shot class incremental learning (FSCIL) method for skin disease diagnosis (SDD) based on the consumer Internet of Things (CIoT) is designed to solve these problems in this paper. First, dermoscopic images and clinical images obtained from CIoT nodes are used to cover more skin disease categories, and a dual-flow modal alignment module is designed to mitigate the modal misalignment of different modal images. Second, a queue of gradient episodic memory (Q-GEM) method is designed to solve the catastrophic forgetting problem. Third, a differential diagnosis method (DDM), which can effectively improve the low classification accuracy of the few-shot learning (FSL) classification network, is designed. Experiments show that the top-3 diagnostic accuracy of the proposed method can match the accuracy level of dermatologists, the accuracy is 11.2% improvement over the SOTA method.


  • CAMRL: A Joint Method of Channel Attention and Multidimensional Regression Loss for 3D Object Detection in Automated Vehicles

    by Honghao Gao, Danqing Fang, Junsheng Xiao, Walayat Hussain, Jung Yoon Kim

    Fully automated vehicles collect information about their road environments to adjust their driving actions, such as braking and slowing down. The development of artificial intelligence (AI) and the Internet of Things (IoT) has improved the cognitive abilities of vehicles, allowing them to detect traffic signs, pedestrians, and obstacles for increasing the intelligence of these transportation systems. Three-dimensional (3D) object detection in front-view images taken by vehicle cameras is important for both object detection and depth estimation. In this paper, a joint channel attention and multidimensional regression loss method for 3D object detection in automated vehicles (called CAMRL) is proposed to improve the average precision of 3D object detection by focusing on the model’s ability to infer the locations and sizes of objects. First, channel attention is introduced to effectively learn the yaw angles from the road images captured by vehicle cameras. Second, a multidimensional regression loss algorithm is designed to further optimize the size and position parameters during the training process. Third, the intrinsic parameters of the camera and the depth estimate of the model are combined to reduce the object depth computation error, allowing us to calculate the distance between an object and the camera after the object’s size is confirmed. As a result, objects are detected, and their depth estimations are validated. Then, the vehicle can determine when and how to stop if an object is nearby. Finally, experiments conducted on the KITTI dataset demonstrate that our method is effective and performs better than other baseline methods, especially in terms of 3D object detection and bird’s-eye view (BEV) evaluation.


  • A Mutually Supervised Graph Attention Network for Few-Shot Segmentation: The Perspective of Fully Utilizing Limited Samples

    by Honghao Gao, Junsheng Xiao, Yuyu Yin, Tong Liu, Jiangang Shi

    Fully supervised semantic segmentation has performed well in many computer vision tasks. However, it is time-consuming because training a model requires a large number of pixel-level annotated samples. Few-shot segmentation has recently become a popular approach to addressing this problem, as it requires only a handful of annotated samples to generalize to new categories. However, the full utilization of limited samples remains an open problem. Thus, in this article, a mutually supervised few-shot segmentation network is proposed. First, the feature maps from intermediate convolution layers are fused to enrich the capacity of feature representation. Second, the support image and query image are combined into a bipartite graph, and the graph attention network is adopted to avoid losing spatial information and increase the number of pixels in the support image to guide the query image segmentation. Third, the attention map of the query image is used as prior information to enhance the support image segmentation, which forms a mutually supervised regime. Finally, the attention maps of the intermediate layers are fused and sent into the graph reasoning layer to infer the pixel categories. Experiments are conducted on the PASCAL VOC- $5^i$ dataset and FSS-1000 dataset, and the results demonstrate the effectiveness and superior performance of our method compared with other baseline methods.


  • Open Access

    JoSDW: Combating Noisy Labels by Dynamic Weight

    by Yaojie Zhang, Huahu Xu, Junsheng Xiao, Minjie Bian

    The real world is full of noisy labels that lead neural networks to perform poorly because deep neural networks (DNNs) are prone to overfitting label noise. Noise label training is a challenging problem relating to weakly supervised learning. The most advanced existing methods mainly adopt a small loss sample selection strategy, such as selecting the small loss part of the sample for network model training. However, the previous literature stopped here, neglecting the performance of the small loss sample selection strategy while training the DNNs, as well as the performance of different stages, and the performance of the collaborative learning of the two networks from disagreement to an agreement, and making a second classification based on this. We train the network using a comparative learning method. Specifically, a small loss sample selection strategy with dynamic weight is designed. This strategy increases the proportion of agreement based on network predictions, gradually reduces the weight of the complex sample, and increases the weight of the pure sample at the same time. A large number of experiments verify the superiority of our method.

  • Open Access

    Mutual Attention Learning Network for Domain Adaptive Person Re-identification

    by Liming Zhu, Junsheng Xiao

    Owing to its important role in video surveillance, person re-identification(re-ID) has become a hot research field of computer vision. The purpose of person re-ID is to identify the same person under different cameras. Domain adaptive person re-ID is to transfer the training model from labeled dataset to unlabeled dataset. Due to the neglect of key channels and key regions, the performance of most existing domain adaptive person re-ID methods is limited. To solve the above problem, this paper proposed a mutual attention learning network model (MALNet). It is based on channel attention and spatial attention mechanism. It consists of two channel-spatial attention convolutional neural networks (channel-spatial attention CNNs). Two networks are pre-trained in a source domain. They adapt to the target domain by the mutual attention learning strategy afterward. They distill and supervise each other to enhance the complementarity of MAL-Net. Moreover, a channel-spatial attention loss (CSAL) function is proposed to constrain and approximate the classification prediction distributions and feature triplet distributions. Extensive comparative experiments are carried out on two large-scale re-ID datasets (Market-1501 and DukeMTMC-reID). The performance of MAL-Net is better than other state-of-the-art methods, which proves the superiority and effectiveness of our method.

  • A Prior-mask-guided Few-shot Learning for Skin Lesion Segmentation

    by Junsheng Xiao, Huahu Xu, Wei Zhao, Chen Cheng, HongHao Gao