Xiaoyu Chai
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
Research Subjects
Xiaoyu Chai has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Hongdou Yao
Institute of Electronic Engineering, National University of Defense Technology, Hefei, Anhui, China
4 shared publications - Jun Chen
National Engineering Research Center for Multimedia Software, School of Computer Science, Wuhan University, Wuhan 430072, China
4 shared publications - Zheng Wang
Wuhan University
3 shared publications - Dongshu Xu
Wuhan University
2 shared publications - Pengfei Han
School of Artificial Intelligence, OPtics and ElectroNics (iOPEN), Northwestern Polytechnical University, Xian, China
2 shared publications - Yansheng Qiu
Wuhan University
2 shared publications
Latest Publications
- Open Access
- 18 March 2026
- Published by Elsevier in SSRN Electronic Journal
VA-MedSAM: Volumetric Adaptive Method for 3D Medical Image Segmentation
In recent years, medical image segmentation has become increasingly important for disease diagnosis and treatment planning.In the field of computer vision,the Segment Anything Model 2 (SAM 2) has set a new standard for segmenting objects in videos. This success has inspired methods like MedSAM-2, which treats 3D medical scans as video sequences to apply SAM 2 to medical data. However, these methods often process scans slice-by-slice, ignoring the important 3D connections between layers. To address this, we propose VA-MedSAM, a volumetric evolution of the MedSAM-2 framework designed to capture 3D context while mitigating the forgetting of spatial features. Specifically, we embed lightweight Volumetric Adapters into the Hiera backbone, utilizing 3D convolutions to efficiently capture short-term inter-slice spatial continuity. To facilitate a smooth integration from pre-trained 2D features to learned 3D representations, we employ a Zero-Initialized Enhanced Gating Mechanism coupled with a Differential Fine-tuning Strategy. Furthermore, to address boundary ambiguity in complex anatomical structures, we introduce a Combined Loss function that synergizes Dice and Binary Cross-Entropy losses, optimizing both region-level overlap and pixel-level classification. Comprehensive experiments are conducted on the BTCV multi-organ dataset and the AMOS22 benchmark. VA-MedSAM achieves state-of-the-art mean Dice scores of 93.21% on BTCV and 95.80% on AMOS22,outperforming the baseline by 1.73% and 2.01%, respectively.
- 18 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessVA-MedSAM: Volumetric Adaptive Method for 3D Medical Image Segmentation
In recent years, medical image segmentation has become increasingly important for disease diagnosis and treatment planning.In the field of computer vision,the Segment Anything Model 2 (SAM 2) has set a new standard for segmenting objects in videos. This success has inspired methods like MedSAM-2, which treats 3D medical scans as video sequences to apply SAM 2 to medical data. However, these methods often process scans slice-by-slice, ignoring the important 3D connections between layers. To address this, we propose VA-MedSAM, a volumetric evolution of the MedSAM-2 framework designed to capture 3D context while mitigating the forgetting of spatial features. Specifically, we embed lightweight Volumetric Adapters into the Hiera backbone, utilizing 3D convolutions to efficiently capture short-term inter-slice spatial continuity. To facilitate a smooth integration from pre-trained 2D features to learned 3D representations, we employ a Zero-Initialized Enhanced Gating Mechanism coupled with a Differential Fine-tuning Strategy. Furthermore, to address boundary ambiguity in complex anatomical structures, we introduce a Combined Loss function that synergizes Dice and Binary Cross-Entropy losses, optimizing both region-level overlap and pixel-level classification. Comprehensive experiments are conducted on the BTCV multi-organ dataset and the AMOS22 benchmark. VA-MedSAM achieves state-of-the-art mean Dice scores of 93.21% on BTCV and 95.80% on AMOS22,outperforming the baseline by 1.73% and 2.01%, respectively.
- 1 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (10) , 19068-19080
MonOri: Orientation-Guided PnP for Monocular 3-D Object Detection
Monocular 3-D object detection is a challenging task in the field of autonomous driving and has made great progress. However, current monocular image methods tend to incorporate additional information such as pseudolabels to improve algorithm performance while overlooking the geometric relationship between the object's keypoints, resulting in low performance for occluded object detection. To address this issue, we find that introducing the orientation information of objects in the 3-D detection pipeline can help improve the detection performance of occluded objects. An orientation-guided perspective-n-point (PnP) for monocular 3-D object detection method named MonOri is presented in this article, which uses object's orientation to guide keypoints' optimization. Considering the existence of different deformation objects in the scene, we design the feature aggregation detection module (FADM), which consists of the feature focus fusion module (FFFM) and CondConv detection module (CCDM). First, FFFM can highlight signals from irregularly occluded objects, effectively modeling features of elongated and small-sized objects. This module enhances the model's ability to recognize elongated and small-sized objects in complex scenes. Then, the CCDM is designed to improve the network's ability to estimate object keypoints' location regression under occlusion conditions and minimize the network computational overhead. Finally, considering that the unoccluded portions of occluded objects are closely related to the orientation of the objects, an orientation-guided keypoints' selection module (OGKSM) is proposed to enhance the accuracy of objected optimization for keypoint positions and spatial location inference of the object. Experimental results indicate that the MonOri method achieves competitive results; it is also demonstrated that the orientation information is introduced in the PnP algorithm to estimate the object's spatial position that can mitigate the impact of occlusion on object detection, thus improving the recognition rate of occluded objects. Our code is available at https://github.com/DL-YHD/MonOri.
- 1 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 36 (10) , 19068-19080
MonOri: Orientation-Guided PnP for Monocular 3-D Object Detection
Monocular 3-D object detection is a challenging task in the field of autonomous driving and has made great progress. However, current monocular image methods tend to incorporate additional information such as pseudolabels to improve algorithm performance while overlooking the geometric relationship between the object's keypoints, resulting in low performance for occluded object detection. To address this issue, we find that introducing the orientation information of objects in the 3-D detection pipeline can help improve the detection performance of occluded objects. An orientation-guided perspective-n-point (PnP) for monocular 3-D object detection method named MonOri is presented in this article, which uses object's orientation to guide keypoints' optimization. Considering the existence of different deformation objects in the scene, we design the feature aggregation detection module (FADM), which consists of the feature focus fusion module (FFFM) and CondConv detection module (CCDM). First, FFFM can highlight signals from irregularly occluded objects, effectively modeling features of elongated and small-sized objects. This module enhances the model's ability to recognize elongated and small-sized objects in complex scenes. Then, the CCDM is designed to improve the network's ability to estimate object keypoints' location regression under occlusion conditions and minimize the network computational overhead. Finally, considering that the unoccluded portions of occluded objects are closely related to the orientation of the objects, an orientation-guided keypoints' selection module (OGKSM) is proposed to enhance the accuracy of objected optimization for keypoint positions and spatial location inference of the object. Experimental results indicate that the MonOri method achieves competitive results; it is also demonstrated that the orientation information is introduced in the PnP algorithm to estimate the object's spatial position that can mitigate the impact of occlusion on object detection, thus improving the recognition rate of occluded objects. Our code is available at https://github.com/DL-YHD/MonOri.
- 20 May 2025
- Published by Elsevier in Pattern Recognition
- Vol. 168
CMANet: A CNN-Mamba aggregation network for face super-resolution
- 20 May 2025
- Published by Elsevier in Pattern Recognition
- Vol. 168
CMANet: A CNN-Mamba aggregation network for face super-resolution
- 29 December 2023
- Published by Springer Nature
- p. 27-39
Multi-image 3D Face Reconstruction via an Adaptive Aggregation Network
- 29 December 2023
- Published by Springer Nature
- p. 27-39
Multi-image 3D Face Reconstruction via an Adaptive Aggregation Network
- 29 December 2023
- Published by Springer Nature
- p. 70-82
An Adaptive-Guidance GAN for Accurate Face Reenactment
- 29 December 2023
- Published by Springer Nature
- p. 70-82
An Adaptive-Guidance GAN for Accurate Face Reenactment
- 18 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (5) , 4593-4605
Occlusion-Aware Plane-Constraints for Monocular 3D Object Detection
The task of 3D object detection poses a significant challenge for 3D scene understanding and is primarily employed in the fields of robot control and autonomous driving. Monocular-based 3D detection methods are more cost-effective and practical than stereo-based or LiDAR-based methods. Monocular image 3D detection methods have garnered considerable attention from researchers. However, the impact of occlusion scenarios of the objects on the keypoints prediction is often overlooked. To address this issue, the present paper proposes a novel 3D monocular object detection method named MonOAPC, which is equipped with occlusion-aware plane-constraints. This method can adaptively utilize partial keypoints to infer the plane location of the object based on the level of occlusion and highlights that the introduction of plane constraints is advantageous for the 3D detection task. First, the plane information of the object in 3D space is beneficial to optimize the keypoints regression, and considering that each plane holds different significance, an adaptive plane location inference module is proposed to enhance the keypoints location regression. Second, a novel co-depth estimation module is proposed to jointly estimate the object’s spatial location through various depth inference methods, thereby improving the generalization of the object depth estimation. Furthermore, the paper demonstrates that the accuracy of 3D object detection can be indirectly improved by introducing plane information to promote keypoints regression, and that plane information is effective for monocular 3D detection. The experimental outcomes show that the MonOAPC method can attain competitive results.
- 18 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Transportation Systems
- Vol. 25 (5) , 4593-4605
Occlusion-Aware Plane-Constraints for Monocular 3D Object Detection
The task of 3D object detection poses a significant challenge for 3D scene understanding and is primarily employed in the fields of robot control and autonomous driving. Monocular-based 3D detection methods are more cost-effective and practical than stereo-based or LiDAR-based methods. Monocular image 3D detection methods have garnered considerable attention from researchers. However, the impact of occlusion scenarios of the objects on the keypoints prediction is often overlooked. To address this issue, the present paper proposes a novel 3D monocular object detection method named MonOAPC, which is equipped with occlusion-aware plane-constraints. This method can adaptively utilize partial keypoints to infer the plane location of the object based on the level of occlusion and highlights that the introduction of plane constraints is advantageous for the 3D detection task. First, the plane information of the object in 3D space is beneficial to optimize the keypoints regression, and considering that each plane holds different significance, an adaptive plane location inference module is proposed to enhance the keypoints location regression. Second, a novel co-depth estimation module is proposed to jointly estimate the object’s spatial location through various depth inference methods, thereby improving the generalization of the object depth estimation. Furthermore, the paper demonstrates that the accuracy of 3D object detection can be indirectly improved by introducing plane information to promote keypoints regression, and that plane information is effective for monocular 3D detection. The experimental outcomes show that the MonOAPC method can attain competitive results.
- 20 May 2013
- Published by American Society of Clinical Oncology (ASCO) in Journal of Clinical Oncology
- Vol. 31 (15_suppl) , 1090
A phase II study evaluating the safety and efficacy of sunitinib with weekly paclitaxel followed by doxorubicin and daily oral cyclophosphamide plus G-CSF as neoadjuvant chemotherapy (NC) for locally advanced (LABC) or inflammatory breast cancer (IBC).
1090 Background: Sunitinib (S) is an oral tyrosine kinase inhibitor with anti-tumor and anti-angiogenic activity. The primary objective of this trial was to assess the pathologic complete response rate (pCR) in patients (pts) treated with NC consisting of S with weekly paclitaxel (T) followed by doxorubicin and daily oral cyclophosphamide plus G-CSF (AC+G-CSF). Correlative studies including circulating biomarkers are reported separately. Methods: Pts with HER2 negative LABC or IBC were eligible for this multicenter, phase II trial. Pts received S 25 mg po daily with T 80 mg/m2 IV Qweek (wk) x 12 wks, then AC+G-CSF (doxorubicin 24 mg/m2 IV Qwk + oral cyclophosphamide 60 mg/m2po daily + G-CSF 5 mcg/kg SC days 2-7) x 15 wks. pCR in the breast and axilla was assessed at surgery, and the MDACC CPS+EG score (validated score combining clinical and pathologic results for predicting survival in the neoadjuvant setting) was calculated. Results: 70 pts (ages 33-79) were enrolled; 68 received protocol therapy. 37 (53%) had ER and/or PR positive tumors. 2 patients were unevaluable (hypersensitivity to T, toxicity possibly related to S) and 3 withdrew consent prior to surgery. 61 pts reported any grade AE in S+P period. Among grade 3 or 4 AEs, neutropenia was most common in S+P period occurring in 31/68 (46%). pCR in the breast was observed in 27% (17/63, 95% CI 17%-40%) and breast and axilla in 15/63 (24%). In pts with ER positive tumors, pCR rate in breast was 8/34 (24%) and 9/29 (31%) for pts with ER negative tumors. 18 evaluable pts (29%) had CPS+EG scores ≤2, 40 (63%) had CPS+EG scores of ≥3, and 5 had insufficient information to calculate the CPS+EG score. When response was defined as pCR and/or CPS+EG score ≤ 2, 28 pts (44%) were responders. Conclusions: NC with T+S followed by AC+G-CSF was well tolerated. Using a combined definition of response of pCR and/or CPS+EG score ≤2, 28/63 (44%) pts had response; 19/34 (56%) for ER positive and 9/29 (31%) for ER negative disease. The addition of S to NC may result in promising incremental benefit for pts with ER positive breast cancer. Clinical trial information: NCT00513695.
- 20 May 2013
- Published by American Society of Clinical Oncology (ASCO) in Journal of Clinical Oncology
- Vol. 31 (15_suppl) , 1090
A phase II study evaluating the safety and efficacy of sunitinib with weekly paclitaxel followed by doxorubicin and daily oral cyclophosphamide plus G-CSF as neoadjuvant chemotherapy (NC) for locally advanced (LABC) or inflammatory breast cancer (IBC).
1090 Background: Sunitinib (S) is an oral tyrosine kinase inhibitor with anti-tumor and anti-angiogenic activity. The primary objective of this trial was to assess the pathologic complete response rate (pCR) in patients (pts) treated with NC consisting of S with weekly paclitaxel (T) followed by doxorubicin and daily oral cyclophosphamide plus G-CSF (AC+G-CSF). Correlative studies including circulating biomarkers are reported separately. Methods: Pts with HER2 negative LABC or IBC were eligible for this multicenter, phase II trial. Pts received S 25 mg po daily with T 80 mg/m2 IV Qweek (wk) x 12 wks, then AC+G-CSF (doxorubicin 24 mg/m2 IV Qwk + oral cyclophosphamide 60 mg/m2po daily + G-CSF 5 mcg/kg SC days 2-7) x 15 wks. pCR in the breast and axilla was assessed at surgery, and the MDACC CPS+EG score (validated score combining clinical and pathologic results for predicting survival in the neoadjuvant setting) was calculated. Results: 70 pts (ages 33-79) were enrolled; 68 received protocol therapy. 37 (53%) had ER and/or PR positive tumors. 2 patients were unevaluable (hypersensitivity to T, toxicity possibly related to S) and 3 withdrew consent prior to surgery. 61 pts reported any grade AE in S+P period. Among grade 3 or 4 AEs, neutropenia was most common in S+P period occurring in 31/68 (46%). pCR in the breast was observed in 27% (17/63, 95% CI 17%-40%) and breast and axilla in 15/63 (24%). In pts with ER positive tumors, pCR rate in breast was 8/34 (24%) and 9/29 (31%) for pts with ER negative tumors. 18 evaluable pts (29%) had CPS+EG scores ≤2, 40 (63%) had CPS+EG scores of ≥3, and 5 had insufficient information to calculate the CPS+EG score. When response was defined as pCR and/or CPS+EG score ≤ 2, 28 pts (44%) were responders. Conclusions: NC with T+S followed by AC+G-CSF was well tolerated. Using a combined definition of response of pCR and/or CPS+EG score ≤2, 28/63 (44%) pts had response; 19/34 (56%) for ER positive and 9/29 (31%) for ER negative disease. The addition of S to NC may result in promising incremental benefit for pts with ER positive breast cancer. Clinical trial information: NCT00513695.