Shaoyun Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shaoyun Xu has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
- Gongyan Li
Institute of Microelectronics, Chinese Academy of Sciences, Beijing 100029, China
5 shared publications - Yuexing Hao
Institute of Microelectronics, Chinese Academy of Sciences, Beijing 100029, China
4 shared publications - Xiaoye Sun
Institute of Microelectronics Chinese Academy of Sciences 100029 Beijing China
3 shared publications Yuexing Hao
Institute of Microelectronics of the Chinese Academy of Sciences, Beijing, China
3 shared publicationsGongyan Li
Institute of Microelectronics of the Chinese Academy of Sciences, Beijing, China
2 shared publications- Chaokun Shi1 shared publication
Latest Publications
- 25 October 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (10) , 670
AGP-MVSNet: diffusion-enhanced adaptive gaussian prior fusion with confidence-aware gating for multi-view stereo
- 25 October 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (10) , 670
AGP-MVSNet: diffusion-enhanced adaptive gaussian prior fusion with confidence-aware gating for multi-view stereo
- Open Access
CPH: Collision-Avoidance-Driven Coordination for Distributed Multi-Agent Path Finding via Multi-Agent Reinforcement Learning
Multi-Agent Reinforcement Learning (MARL) provides an effective paradigm for distributed Multi-Agent Path Finding (MAPF) under partial observability. Due to frequent local interactions among agents, action decisions based on limited local information can induce potential collision risks and may further lead to action conflicts, posing challenges to efficient multi-agent coordination. In this paper, we propose CPH, a value-based MARL framework that proactively mitigates collision risks during path decision-making and resolves unavoidable action conflicts before execution. Specifically, we introduce a collision-risk-driven communication mechanism that selects partners according to potential collision relationships and integrates collision-aware spatial contexts with agent-centric observations via cross-attention. Moreover, we propose a potential-inspired aggregation reward shaping method, which encourages agents to disperse from collision-prone aggregation states. Additionally, we present a conflict-aware hybrid post-processing strategy that adaptively resolves different conflict scenarios through local action adjustment. Experimental results show that CPH outperforms the compared distributed MARL-based MAPF methods, while ablation studies confirm the effectiveness of all three components. These results indicate that combining proactive collision-risk mitigation with adaptive conflict resolution before execution improves path reachability and planning efficiency in distributed MAPF.
Open AccessCPH: Collision-Avoidance-Driven Coordination for Distributed Multi-Agent Path Finding via Multi-Agent Reinforcement Learning
Multi-Agent Reinforcement Learning (MARL) provides an effective paradigm for distributed Multi-Agent Path Finding (MAPF) under partial observability. Due to frequent local interactions among agents, action decisions based on limited local information can induce potential collision risks and may further lead to action conflicts, posing challenges to efficient multi-agent coordination. In this paper, we propose CPH, a value-based MARL framework that proactively mitigates collision risks during path decision-making and resolves unavoidable action conflicts before execution. Specifically, we introduce a collision-risk-driven communication mechanism that selects partners according to potential collision relationships and integrates collision-aware spatial contexts with agent-centric observations via cross-attention. Moreover, we propose a potential-inspired aggregation reward shaping method, which encourages agents to disperse from collision-prone aggregation states. Additionally, we present a conflict-aware hybrid post-processing strategy that adaptively resolves different conflict scenarios through local action adjustment. Experimental results show that CPH outperforms the compared distributed MARL-based MAPF methods, while ablation studies confirm the effectiveness of all three components. These results indicate that combining proactive collision-risk mitigation with adaptive conflict resolution before execution improves path reachability and planning efficiency in distributed MAPF.
- 2 March 2025
- Published by Springer Nature
- p. 520-531
Multi-agent Path Finding Based on Improved CBS Algorithm
- 2 March 2025
- Published by Springer Nature
- p. 520-531
Multi-agent Path Finding Based on Improved CBS Algorithm
- Open Access
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 158-166
Optimization Strategy for MAPF Problem with Traffic Rule-based CBS Algorithm in Warehouses
- 22 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 158-166
Open AccessOptimization Strategy for MAPF Problem with Traffic Rule-based CBS Algorithm in Warehouses
- 7 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 257
Knowledge distillation via Noisy Feature Reconstruction
- 7 August 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 257
Knowledge distillation via Noisy Feature Reconstruction
- 10 April 2023
- Published by Springer Nature in Neural Computing and Applications
- Vol. 35 (21) , 1-15
CoG-Trans: coupled graph convolutional transformer for multi-label classification of cherry defects
- 10 April 2023
- Published by Springer Nature in Neural Computing and Applications
- Vol. 35 (21) , 1-15
CoG-Trans: coupled graph convolutional transformer for multi-label classification of cherry defects
- Open Access
Implementation of Lightweight Convolutional Neural Networks via Layer-Wise Differentiable Compression
Convolutional neural networks (CNNs) have achieved significant breakthroughs in various domains, such as natural language processing (NLP), and computer vision. However, performance improvement is often accompanied by large model size and computation costs, which make it not suitable for resource-constrained devices. Consequently, there is an urgent need to compress CNNs, so as to reduce model size and computation costs. This paper proposes a layer-wise differentiable compression (LWDC) algorithm for compressing CNNs structurally. A differentiable selection operator OS is embedded in the model to compress and train the model simultaneously by gradient descent in one go. Instead of pruning parameters from redundant operators by contrast to most of the existing methods, our method replaces the original bulky operators with more lightweight ones directly, which only needs to specify the set of lightweight operators and the regularization factor in advance, rather than the compression rate for each layer. The compressed model produced by our method is generic and does not need any special hardware/software support. Experimental results on CIFAR-10, CIFAR-100 and ImageNet have demonstrated the effectiveness of our method. LWDC obtains more significant compression than state-of-the-art methods in most cases, while having lower performance degradation. The impact of lightweight operators and regularization factor on the compression rate and accuracy also is evaluated.
Open AccessImplementation of Lightweight Convolutional Neural Networks via Layer-Wise Differentiable Compression
Convolutional neural networks (CNNs) have achieved significant breakthroughs in various domains, such as natural language processing (NLP), and computer vision. However, performance improvement is often accompanied by large model size and computation costs, which make it not suitable for resource-constrained devices. Consequently, there is an urgent need to compress CNNs, so as to reduce model size and computation costs. This paper proposes a layer-wise differentiable compression (LWDC) algorithm for compressing CNNs structurally. A differentiable selection operator OS is embedded in the model to compress and train the model simultaneously by gradient descent in one go. Instead of pruning parameters from redundant operators by contrast to most of the existing methods, our method replaces the original bulky operators with more lightweight ones directly, which only needs to specify the set of lightweight operators and the regularization factor in advance, rather than the compression rate for each layer. The compressed model produced by our method is generic and does not need any special hardware/software support. Experimental results on CIFAR-10, CIFAR-100 and ImageNet have demonstrated the effectiveness of our method. LWDC obtains more significant compression than state-of-the-art methods in most cases, while having lower performance degradation. The impact of lightweight operators and regularization factor on the compression rate and accuracy also is evaluated.
- 24 October 2020
- Published by Springer Nature in Machine Vision and Applications
- Vol. 32 (1) , 1-13
FSD: feature skyscraper detector for stem end and blossom end of navel orange
- 24 October 2020
- Published by Springer Nature in Machine Vision and Applications
- Vol. 32 (1) , 1-13
FSD: feature skyscraper detector for stem end and blossom end of navel orange
- Open Access
Fastidious Attention Network for Navel Orange Segmentation
Deep learning achieves excellent performance in many domains, so we not only apply it to the navel orange semantic segmentation task to solve the two problems of distinguishing defect categories and identifying the stem end and blossom end, but also propose a fastidious attention mechanism to further improve model performance. This lightweight attention mechanism includes two learnable parameters, activations and thresholds, to capture long-range dependence. Specifically, the threshold picks out part of the spatial feature map and the activation excite this area. Based on activations and thresholds training from different types of feature maps, we design fastidious self-attention module (FSAM) and fastidious inter-attention module (FIAM). And then construct the Fastidious Attention Network (FANet), which uses U-Net as the backbone and embeds these two modules, to solve the problems with semantic segmentation for stem end, blossom end, flaw and ulcer. Compared with some state-of-the-art deep-learning-based networks under our navel orange dataset, experiments show that our network is the best performance with pixel accuracy 99.105%, mean accuracy 77.468%, mean IU 70.375% and frequency weighted IU 98.335%. And embedded modules show better discrimination of 5 categories including background, especially the IU of flaw is increased by 3.165%.
Open AccessFastidious Attention Network for Navel Orange Segmentation
Deep learning achieves excellent performance in many domains, so we not only apply it to the navel orange semantic segmentation task to solve the two problems of distinguishing defect categories and identifying the stem end and blossom end, but also propose a fastidious attention mechanism to further improve model performance. This lightweight attention mechanism includes two learnable parameters, activations and thresholds, to capture long-range dependence. Specifically, the threshold picks out part of the spatial feature map and the activation excite this area. Based on activations and thresholds training from different types of feature maps, we design fastidious self-attention module (FSAM) and fastidious inter-attention module (FIAM). And then construct the Fastidious Attention Network (FANet), which uses U-Net as the backbone and embeds these two modules, to solve the problems with semantic segmentation for stem end, blossom end, flaw and ulcer. Compared with some state-of-the-art deep-learning-based networks under our navel orange dataset, experiments show that our network is the best performance with pixel accuracy 99.105%, mean accuracy 77.468%, mean IU 70.375% and frequency weighted IU 98.335%. And embedded modules show better discrimination of 5 categories including background, especially the IU of flaw is increased by 3.165%.
- Open Access
FSD: Feature Skyscraper Detector for Stem End and Blossom End of Navel Orange
To accurately and efficiently distinguish the stem end and the blossom end of navel orange from its black spots, we propose a feature skyscraper detector (FSD) with low computational cost, compact architecture and high detection accuracy. The main part of the detector is inspired from small object that stem (blossom) end is complex and black spot is densely distributed, so we design the feature skyscraper networks (FSN) based on dense connectivity. In particular, FSN is distinguished from regular feature pyramids, and which provides more intensive detection of high-level features. Then we design the backbone of the FSD based on attention mechanism and dense block for better feature extraction to the FSN. In addition, the architecture of the detector is also added Swish to further improve the accuracy. And we create a dataset in Pascal VOC format annotated three types of detection targets the stem end, the blossom end and the black spot. Experimental results on our orange data set confirm that FSD has competitive results to the state-of-the-art one-stage detectors like SSD, DSOD, YOLOv2, YOLOv3, RFB and FSSD, and it achieves 87.479%mAP at 131 FPS with only 5.812M parameters.
Open AccessFSD: Feature Skyscraper Detector for Stem End and Blossom End of Navel Orange
To accurately and efficiently distinguish the stem end and the blossom end of navel orange from its black spots, we propose a feature skyscraper detector (FSD) with low computational cost, compact architecture and high detection accuracy. The main part of the detector is inspired from small object that stem (blossom) end is complex and black spot is densely distributed, so we design the feature skyscraper networks (FSN) based on dense connectivity. In particular, FSN is distinguished from regular feature pyramids, and which provides more intensive detection of high-level features. Then we design the backbone of the FSD based on attention mechanism and dense block for better feature extraction to the FSN. In addition, the architecture of the detector is also added Swish to further improve the accuracy. And we create a dataset in Pascal VOC format annotated three types of detection targets the stem end, the blossom end and the black spot. Experimental results on our orange data set confirm that FSD has competitive results to the state-of-the-art one-stage detectors like SSD, DSOD, YOLOv2, YOLOv3, RFB and FSSD, and it achieves 87.479%mAP at 131 FPS with only 5.812M parameters.