Xichun Sheng

Xichun Sheng

Auto-generated profile, based on publicly available information

Xichun Sheng

Macao Polytechnic University, Macao, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
3 December 2025 in IEEE Transactions on Multimedia
Verified
Journal Article
Closed access
BGBoyang Guo
Boyang Guo
Liang LiLiang Li
Liang Li
JZJiehua Zhang
Jiehua Zhang
YSYaoqi Sun
Yaoqi Sun
CYChenggang Yan
Chenggang Yan
XSXichun Sheng
Xichun Sheng
Prompt learning is an effective way to adapt pre-trained models to downstream tasks by training a small number of additional learnable prompts. Recent studies address several early challenges by combining generalized knowledge from frozen pre-trained VL models with task-specific knowledge from training data as guidance for prompt learning. However, existing methods still struggle with the generalization-adaptation (GA) trade-off dilemma: excessive reliance on generalized knowledge hinders adaptation to downstream tasks, while overemphasis on task-specific knowledge undermines the inherent generalization capabilities of pre-trained models. To address this issue, we propose a novel prompt learning method called Prompt Learning with Knowledge Regularization (PLKR). PLKR effectively mitigates the GA trade-off dilemma by offering greater flexibility in adapting to task-specific knowledge while minimizing the disruption of pre-trained knowledge. Specifically, we propose category-invariant and topology-invariant knowledge regularization to preserve generalized knowledge: the former enhances category-level discriminative capabilities while allowing flexible task-specific learning, and the latter maintains global topological stability during adaptation to new tasks. Through the proposed regularization, PLKR improves the performance on both base and new tasks. We evaluate the effectiveness of our approach on four representative tasks over 11 datasets. Experimental results show our method outperforms existing SOTA methods by a large margin.

23 June 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
JZJinkai Zheng
Jinkai Zheng
MLMinhao Lin
Minhao Lin
JLJiaBin Li
JiaBin Li
YZYanting Zhang
Yanting Zhang
SZShaoxiong Zhang
Shaoxiong Zhang
XLXinchen Liu
Xinchen Liu
Existing studies of gait recognition mainly use binary silhouette, human parsing, or skeleton to encode the shape and dynamics of the human body during walking. Although they have achieved certain success, they still face significant bottlenecks in temporal modeling, especially in in-the-wild scenarios where irregular walking patterns challenge the extraction of stable temporal features. In contrast, optical flow, as a rising gait modality with strong temporal characteristics, is inherently better at capturing motion dynamics. To unleash the power of optical flow in gait recognition in the wild, this paper proposes a novel optical flow-driven multimodal integration gait recognition framework, named FlowGait. The FlowGait consists of the Global Attention Module (GAM), the Part-aware Attention Module (PAM), and the Cross-granularity Alignment Module (CAM). To obtain complementary spatial-temporal information, the GAM and PAM are elaborately designed to integrate temporal features with global and local spatial features, respectively. Then, the CAM achieves fine-grained alignment of spatial-temporal features at the high level, thereby enhancing the generalization ability of the model. Moreover, to make up for the shortage of data, we build a large-scale optical flow-based dataset for gait recognition in the wild, named Gait3D-Flow. Extensive experiments on the Gait3D, CCPG, and SUSTech1K datasets demonstrate the superiority of our FlowGait, with the Rank-1 accuracy of 87.3%, 95.8%, and 88.2%, respectively. The code and dataset will be released for research purposes.

24 July 2026 in Journal of Visual Communication and Image Representation
Verified
Journal Article
Closed access
XSXichun Sheng
Xichun Sheng
Gaopeng HuangGaopeng Huang
Gaopeng Huang
TTTao Tan
Tao Tan
BWBin Wan
Bin Wan
GZGuangyuan Zhang
Guangyuan Zhang
LSLihuan Shao
Lihuan Shao

30 April 2026 in Journal of Visual Communication and Image Representation
Verified
Journal Article
Open access
XSXichun Sheng
Xichun Sheng
YSYaoqi Sun
Yaoqi Sun
TTTao Tan
Tao Tan
ZLZhihao Li
Zhihao Li
ZHZhao Huang
Zhao Huang
XHXingru Huang
Xingru Huang