Wei-Jong Yang
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
Wei-Jong Yang has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Jar-Ferr Yang
System-on-Chip Research Center, National Cheng Kung University, Tainan 701401, Taiwan
6 shared publications - Pau-Choo Chung
Department of Electrical Engineering, National Cheng Kung University, Tainan, Taiwan
5 shared publications - Din-Yuen Chan
Department of Computer Science and Information Engineering, National Chiayi University, Chiayi 600355, Taiwan
4 shared publications - Hung-Wen Tsai
Department of Pathology, National Cheng Kung University Hospital, Tainan, Taiwan
3 shared publications - Che-Wei Hsu
Department of Pathology, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan
2 shared publications - Hsuan Tsai
Institute of Computer and Communication Engineering, Department of Electrical Engineering, National Cheng Kung University, Tainan 701401, Taiwan
2 shared publications
Latest Publications
- Open Access
- 29 October 2026
- Published by Elsevier in Neurocomputing
- Vol. 711
CroSSFed: Prototype-guided server-side finetuning with client-server reweighting for cross-domain semi-supervised federated learning
- 29 October 2026
- Published by Elsevier in Neurocomputing
- Vol. 711
Open AccessCroSSFed: Prototype-guided server-side finetuning with client-server reweighting for cross-domain semi-supervised federated learning
- Open Access
- 11 August 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (16) , 8003
Improved Classification-Based Monocular Depth Estimation for Robust Generalization of Diverse Scenes
Recently, monocular depth estimation methods with neural networks trained using an image-depth database have become widely used and prevalent techniques. However, their generalization abilities are often constrained by training data: models trained on indoor datasets may perform well on other indoor datasets but exhibit some performance degradation for outdoor scenes. Therefore, developing a monocular depth estimation network with strong generalization capability for diverse scenes is challenging. In this paper, we propose a monocular depth estimation model that exhibits high generalizability and performs effectively across various scenarios without additional fine-tuning. In terms of model design, we adopt a powerful visual encoder to extract rich semantic features and incorporate the proposed multiple depth prediction heads with self-adaptability for robust depth estimation. With the designed modules and architectural refinements, the proposed model achieves a favorable balance between parameter efficiency and inference accuracy. After testing several unseen datasets, the simulation results demonstrate that the proposed method achieves better estimation performance than the baseline method and shows better robust visualization performance on multiple seen and unseen datasets.
- 11 August 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (16) , 8003
Open AccessImproved Classification-Based Monocular Depth Estimation for Robust Generalization of Diverse Scenes
Recently, monocular depth estimation methods with neural networks trained using an image-depth database have become widely used and prevalent techniques. However, their generalization abilities are often constrained by training data: models trained on indoor datasets may perform well on other indoor datasets but exhibit some performance degradation for outdoor scenes. Therefore, developing a monocular depth estimation network with strong generalization capability for diverse scenes is challenging. In this paper, we propose a monocular depth estimation model that exhibits high generalizability and performs effectively across various scenarios without additional fine-tuning. In terms of model design, we adopt a powerful visual encoder to extract rich semantic features and incorporate the proposed multiple depth prediction heads with self-adaptability for robust depth estimation. With the designed modules and architectural refinements, the proposed model achieves a favorable balance between parameter efficiency and inference accuracy. After testing several unseen datasets, the simulation results demonstrate that the proposed method achieves better estimation performance than the baseline method and shows better robust visualization performance on multiple seen and unseen datasets.
- Open Access
Multiscale Region-Based Convolutional Neural Networks for 3D Object Detection with LiDAR Sensors
LiDAR-based 3D object detection is essential for autonomous driving vehicles under poor lighting conditions. With LiDAR data, point cloud technologies have become increasingly important, as LiDAR sensors are largely cost down. However, the sparsity of point cloud poses a challenge for 3D object detection, requiring advancements in sparse convolutional networks. Given that the multiscale feature fusion mechanism can improve object detection performance using rich information across scale features, we added a refinement fusion network with cross-attention modules to existing 3D voxel-based object detection networks. We also employed a realistic strategy to refine existing point cloud data augmentation techniques to enable the trained detection networks to achieve substantially improved results. The experimental results demonstrate the effectiveness of our proposed detection system across three categories on the KITTI dataset. These enhancements address the limitations of current approaches and highlight the superior performance of the proposed system.
Open AccessMultiscale Region-Based Convolutional Neural Networks for 3D Object Detection with LiDAR Sensors
LiDAR-based 3D object detection is essential for autonomous driving vehicles under poor lighting conditions. With LiDAR data, point cloud technologies have become increasingly important, as LiDAR sensors are largely cost down. However, the sparsity of point cloud poses a challenge for 3D object detection, requiring advancements in sparse convolutional networks. Given that the multiscale feature fusion mechanism can improve object detection performance using rich information across scale features, we added a refinement fusion network with cross-attention modules to existing 3D voxel-based object detection networks. We also employed a realistic strategy to refine existing point cloud data augmentation techniques to enable the trained detection networks to achieve substantially improved results. The experimental results demonstrate the effectiveness of our proposed detection system across three categories on the KITTI dataset. These enhancements address the limitations of current approaches and highlight the superior performance of the proposed system.
- Open Access
- 13 January 2026
- Published by MDPI AG in Preprints.org
Multiscale Region-Based Convolutional Neural Networks for 3D Object Detection with LiDAR Sensors
LiDAR-based 3D object detection plays an essential role for autonomous driving vehicles under poor lighting condition environments. With LiDAR data, the significance of point cloud technologies is increasing important since Li-DAR sensors are largely cost down. However, the sparsity of point cloud poses a challenge for 3D object detection such that the advancements of sparse convolutional networks could be further investigated. Considering multiscale feature fusion mechanism can improve the object detection performances by using the rich inform among various scale features, we properly add the refinement fusion network with cross-attention modules to the existed 3D voxel-based object detection networks. Additionally, we a realistic strategy to refine the existing point-cloud data augmentation techniques to enable the trained detection networks to achieve substantially improved results. The experimental results demonstrate the effectiveness of our proposed detection system across three categories on the KITTI dataset. These enhancements address the limitations of current approaches and highlight the superior performance of the proposed system.
- 13 January 2026
- Published by MDPI AG in Preprints.org
Open AccessMultiscale Region-Based Convolutional Neural Networks for 3D Object Detection with LiDAR Sensors
LiDAR-based 3D object detection plays an essential role for autonomous driving vehicles under poor lighting condition environments. With LiDAR data, the significance of point cloud technologies is increasing important since Li-DAR sensors are largely cost down. However, the sparsity of point cloud poses a challenge for 3D object detection such that the advancements of sparse convolutional networks could be further investigated. Considering multiscale feature fusion mechanism can improve the object detection performances by using the rich inform among various scale features, we properly add the refinement fusion network with cross-attention modules to the existed 3D voxel-based object detection networks. Additionally, we a realistic strategy to refine the existing point-cloud data augmentation techniques to enable the trained detection networks to achieve substantially improved results. The experimental results demonstrate the effectiveness of our proposed detection system across three categories on the KITTI dataset. These enhancements address the limitations of current approaches and highlight the superior performance of the proposed system.
- Open Access
- 30 May 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (11) , 6169
High-Precision Depth Estimation Networks Using Low-Resolution Depth and RGB Image Sensors for Low-Cost Mixed Reality Glasses
In recent years, with the booming development of the three-dimensional and mixed reality (MR) industries, depth estimation tools have become increasingly important to support many visual problems. Intrinsically, the restriction of computational resources hardly lets complex depth completion methods be implemented on MR glasses. In this paper, we propose a competitive high-precision depth estimation network, which integrates a dual-path autoencoder and adaptive bin depth estimator together. The proposed network with different types of models can aptly fuse a pair of an RGB image and low-resolution low-quality depth map to generate a high-precision depth map. The simplest model of the proposed network has only 2.28 M weights and 1.150 G MACs. It is very lightweight, such that this model could be easily implemented into the platform of low-cost MR glasses to support hand-gesture controls realized in edge devices in real time.
- 30 May 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (11) , 6169
Open AccessHigh-Precision Depth Estimation Networks Using Low-Resolution Depth and RGB Image Sensors for Low-Cost Mixed Reality Glasses
In recent years, with the booming development of the three-dimensional and mixed reality (MR) industries, depth estimation tools have become increasingly important to support many visual problems. Intrinsically, the restriction of computational resources hardly lets complex depth completion methods be implemented on MR glasses. In this paper, we propose a competitive high-precision depth estimation network, which integrates a dual-path autoencoder and adaptive bin depth estimator together. The proposed network with different types of models can aptly fuse a pair of an RGB image and low-resolution low-quality depth map to generate a high-precision depth map. The simplest model of the proposed network has only 2.28 M weights and 1.150 G MACs. It is very lightweight, such that this model could be easily implemented into the platform of low-cost MR glasses to support hand-gesture controls realized in edge devices in real time.
- Open Access
- 26 April 2025
- Published by MDPI AG in Preprints.org
High-Precision Depth Estimation Networks Using Low-Resolution Depth and RGB Image Sensors for Low Cost MR Glasses
In recent years, with the booming development of the three-dimensional and mixed reality (MR) industries, the depth estimation tool has become increasingly important to support many visual problems. Intrinsically, the restriction of computational resource lets the complex depth completion methods be hardly implemented on MR glasses. In this paper, we propose a competitive high-precision depth estimation network, which integrates dual-path autoencoder and adaptive bin depth estimator together. Thus, the proposed network with different types of models can successfully fuse the high-resolution RGB information and low-resolution depth sensor data to achieve high-accuracy depth estimation. The simplest type of the proposed lightweight models can be implemented into the platform of low-cost MR glasses to support the edge-computing of hand-gesture controls online.
- 26 April 2025
- Published by MDPI AG in Preprints.org
Open AccessHigh-Precision Depth Estimation Networks Using Low-Resolution Depth and RGB Image Sensors for Low Cost MR Glasses
In recent years, with the booming development of the three-dimensional and mixed reality (MR) industries, the depth estimation tool has become increasingly important to support many visual problems. Intrinsically, the restriction of computational resource lets the complex depth completion methods be hardly implemented on MR glasses. In this paper, we propose a competitive high-precision depth estimation network, which integrates dual-path autoencoder and adaptive bin depth estimator together. Thus, the proposed network with different types of models can successfully fuse the high-resolution RGB information and low-resolution depth sensor data to achieve high-accuracy depth estimation. The simplest type of the proposed lightweight models can be implemented into the platform of low-cost MR glasses to support the edge-computing of hand-gesture controls online.
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (8) , 5912-5921
BSMatch: Boundary Segmentation and Matching for Lipid Droplet Quantification in Diagnosis of Non-Alcoholic Fatty Liver Disease
Hepatic steatosis is one of the most obvious indicators of nonalcoholic fatty liver disease. However, the presence of many regions with a similar color and shape as lipid droplets in the histopathological image complicates the task of detecting genuine lipid droplets using automated methods. Accordingly, the present study proposes a boundary segmentation and matching (BSMatch) algorithm for the segmentation of lipid droplets based on their unique boundary characteristics. A two-branch RnB-Unet model is trained to segment the regions and boundaries of the droplets, respectively, in accordance with a boundary matching (BM) loss which enforces the consistency between them. A boundary matching score (BMS) measure is then used to improve the precision of the instance segmentation evaluation process by discarding segmented regions which are not well-matched with their predicted boundaries. The experimental results obtained using a H&E-stained liver slide dataset show that BSMatch outperforms existing methods in terms of both the IoU and the F1-score. The BSMatch results are used to predict the fat percentage in hepatocytes (FPH) in liver whole slide images. The predicted FPH values are well correlated with the steatosis grades assigned by experienced pathologists. Thus, BSMatch appears to have significant promise for NAFLD diagnosis in clinical contexts.
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (8) , 5912-5921
BSMatch: Boundary Segmentation and Matching for Lipid Droplet Quantification in Diagnosis of Non-Alcoholic Fatty Liver Disease
Hepatic steatosis is one of the most obvious indicators of nonalcoholic fatty liver disease. However, the presence of many regions with a similar color and shape as lipid droplets in the histopathological image complicates the task of detecting genuine lipid droplets using automated methods. Accordingly, the present study proposes a boundary segmentation and matching (BSMatch) algorithm for the segmentation of lipid droplets based on their unique boundary characteristics. A two-branch RnB-Unet model is trained to segment the regions and boundaries of the droplets, respectively, in accordance with a boundary matching (BM) loss which enforces the consistency between them. A boundary matching score (BMS) measure is then used to improve the precision of the instance segmentation evaluation process by discarding segmented regions which are not well-matched with their predicted boundaries. The experimental results obtained using a H&E-stained liver slide dataset show that BSMatch outperforms existing methods in terms of both the IoU and the F1-score. The BSMatch results are used to predict the fat percentage in hepatocytes (FPH) in liver whole slide images. The predicted FPH values are well correlated with the steatosis grades assigned by experienced pathologists. Thus, BSMatch appears to have significant promise for NAFLD diagnosis in clinical contexts.
- Open Access
- 19 November 2024
- Published by MDPI AG in Preprints.org
RVTAF: Residual Vision Transformer and Adaptive Fusion Autoencoders for Monocular Depth Estimation
Precision depth estimation plays a key role in many applications, including 3D scene reconstruction, virtual reality, autonomous driving and human-computer interaction. Recent advancements in deep learning technologies, the monocular depth estimation has surpassed the traditional stereo camera systems, bringing new possibilities in 3D sensing. In this paper, by using single camera, we propose an end-to-end supervised monocular depth estimation autoencoder which contains a CNN-ViT encoder and an adaptive fusion decoder, to obtain high-precision depth maps. In the CNN-ViT encoder, we construct a multi-scale feature extractor by mixing residual configurations of vision transformers to enhance both local and global information. In the adaptive fusion decoder, we introduce adaptive fusion modules to effectively merge features of the encoder and decoder together. Lastly, the model is trained using a loss function that aligns with human perception to enable it to focus on the depth values of foreground objects. Experimental results demonstrate the effective prediction of the depth map from a single-view color image by the proposed RVTAF autoencoder.
- 19 November 2024
- Published by MDPI AG in Preprints.org
Open AccessRVTAF: Residual Vision Transformer and Adaptive Fusion Autoencoders for Monocular Depth Estimation
Precision depth estimation plays a key role in many applications, including 3D scene reconstruction, virtual reality, autonomous driving and human-computer interaction. Recent advancements in deep learning technologies, the monocular depth estimation has surpassed the traditional stereo camera systems, bringing new possibilities in 3D sensing. In this paper, by using single camera, we propose an end-to-end supervised monocular depth estimation autoencoder which contains a CNN-ViT encoder and an adaptive fusion decoder, to obtain high-precision depth maps. In the CNN-ViT encoder, we construct a multi-scale feature extractor by mixing residual configurations of vision transformers to enhance both local and global information. In the adaptive fusion decoder, we introduce adaptive fusion modules to effectively merge features of the encoder and decoder together. Lastly, the model is trained using a loss function that aligns with human perception to enable it to focus on the depth values of foreground objects. Experimental results demonstrate the effective prediction of the depth map from a single-view color image by the proposed RVTAF autoencoder.
- Open Access
- 20 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Engineering in Medicine and Biology
- Vol. 5, 261-270
Lymphocyte-Infiltrated Periportal Region Detection With Structurally-Refined Deep Portal Segmentation and Heterogeneous Infiltration Features
Goal : The early diagnosis and treatment of hepatitis is essential to reduce hepatitis-related liver function deterioration and mortality. One component of the widely-used Ishak grading system for the grading of periportal interface hepatitis is based on the percentage of portal borders infiltrated by lymphocytes. Thus, the accurate detection of lymphocyte-infiltrated periportal regions is critical in the diagnosis of hepatitis. However, the infiltrating lymphocytes usually result in the formation of ambiguous and highly-irregular portal boundaries, and thus identifying the infiltrated portal boundary regions precisely using automated methods is challenging. This study aims to develop a deep-learning-based automatic detection framework to assist diagnosis. Methods : The present study proposes a framework consisting of a Structurally-REfined Deep Portal Segmentation module and an Infiltrated Periportal Region Detection module based on heterogeneous infiltration features to accurately identify the infiltrated periportal regions in liver Whole Slide Images. Results : The proposed method achieves 0.725 in F1-score of lymphocyte-infiltrated periportal region detection. Moreover, the statistics of the ratio of the detected infiltrated portal boundary have high correlation to the Ishak grade (Spearman's correlations more than 0.87 with p-values less than 0.001) and medium correlation to the liver function index aspartate aminotransferase and alanine aminotransferase (Spearman's correlations more than 0.63 and 0.57 with p-values less than 0.001). Conclusions : The study shows the statistics of the ratio of infiltrated portal boundary have correlation to the Ishak grade and liver function index. The proposed framework provides pathologists with a useful and reliable tool for hepatitis diagnosis.
- 20 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Engineering in Medicine and Biology
- Vol. 5, 261-270
Open AccessLymphocyte-Infiltrated Periportal Region Detection With Structurally-Refined Deep Portal Segmentation and Heterogeneous Infiltration Features
Goal : The early diagnosis and treatment of hepatitis is essential to reduce hepatitis-related liver function deterioration and mortality. One component of the widely-used Ishak grading system for the grading of periportal interface hepatitis is based on the percentage of portal borders infiltrated by lymphocytes. Thus, the accurate detection of lymphocyte-infiltrated periportal regions is critical in the diagnosis of hepatitis. However, the infiltrating lymphocytes usually result in the formation of ambiguous and highly-irregular portal boundaries, and thus identifying the infiltrated portal boundary regions precisely using automated methods is challenging. This study aims to develop a deep-learning-based automatic detection framework to assist diagnosis. Methods : The present study proposes a framework consisting of a Structurally-REfined Deep Portal Segmentation module and an Infiltrated Periportal Region Detection module based on heterogeneous infiltration features to accurately identify the infiltrated periportal regions in liver Whole Slide Images. Results : The proposed method achieves 0.725 in F1-score of lymphocyte-infiltrated periportal region detection. Moreover, the statistics of the ratio of the detected infiltrated portal boundary have high correlation to the Ishak grade (Spearman's correlations more than 0.87 with p-values less than 0.001) and medium correlation to the liver function index aspartate aminotransferase and alanine aminotransferase (Spearman's correlations more than 0.63 and 0.57 with p-values less than 0.001). Conclusions : The study shows the statistics of the ratio of infiltrated portal boundary have correlation to the Ishak grade and liver function index. The proposed framework provides pathologists with a useful and reliable tool for hepatitis diagnosis.
- 16 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (2) , 613-623
Video-Based Depth Estimation Autoencoder With Weighted Temporal Feature and Spatial Edge Guided Modules
Convolutional neural networks with encoder and decoder structures, generally referred to as autoencoders, are used in many pixel- wise transformation, detection, segmentation and estimation applications, for example, which can be applied for face swapping, lane detection, semantic segmentation, and depth estimation, respectively. However, traditional autoencoders, which are based on single-frame inputs, ignore the temporal consistency between consecutive frames, and may hence produce unsatisfactory results. Accordingly, in this paper, a video-based depth estimation (VDE) autoencoder is proposed to improve the quality of depth estimation through the inclusion of two weighted temporal feature (WTF) modules in the encoder and a single spatial edge guided (SEG) module in the decoder. The WTF modules designed with channel weighted block (CWB) submodule effectively extract the temporal similarities in consecutive frames, while the SEG module provides spatial edge guidance of the object contours. Through the collaboration of the proposed modules, the accuracy of the depth estimation is greatly improved. The experimental results confirm that the proposed VDE autoencoder achieves a better monocular depth estimation performance than the existing autoencoders with only a slight increase in the computational cost.
- 16 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 5 (2) , 613-623
Video-Based Depth Estimation Autoencoder With Weighted Temporal Feature and Spatial Edge Guided Modules
Convolutional neural networks with encoder and decoder structures, generally referred to as autoencoders, are used in many pixel- wise transformation, detection, segmentation and estimation applications, for example, which can be applied for face swapping, lane detection, semantic segmentation, and depth estimation, respectively. However, traditional autoencoders, which are based on single-frame inputs, ignore the temporal consistency between consecutive frames, and may hence produce unsatisfactory results. Accordingly, in this paper, a video-based depth estimation (VDE) autoencoder is proposed to improve the quality of depth estimation through the inclusion of two weighted temporal feature (WTF) modules in the encoder and a single spatial edge guided (SEG) module in the decoder. The WTF modules designed with channel weighted block (CWB) submodule effectively extract the temporal similarities in consecutive frames, while the SEG module provides spatial edge guidance of the object contours. Through the collaboration of the proposed modules, the accuracy of the depth estimation is greatly improved. The experimental results confirm that the proposed VDE autoencoder achieves a better monocular depth estimation performance than the existing autoencoders with only a slight increase in the computational cost.