Zhaotin Yin

Mr. Zhaotin Yin


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
28 April 2025 in IEEE Signal Processing Letters
Verified
Journal Article
Closed access
ZYZhaotin Yin
Zhaotin Yin
HCHaipeng Chen
Haipeng Chen
ZLZhenguang Liu
Zhenguang Liu
SHSihao Hu
Sihao Hu
YJYingying Jiao
Yingying Jiao
Skeleton-based action recognition is crucial for machine intelligence. Current methods generally learn from 3D articulated motion sequences in the straightforward Euclidean space. Yet, the vanilla Euclidean space may not be the optimal choice for modeling the intricate correlations among human body joints. This challenge arises from the non-Euclidean nature of human anatomy, where joint correlations often vary non-linearly during movement. To address this, we propose a dual space representation learning method. Specifically, we represent the motion sequences in Hyperbolic space, leveraging its intrinsic properties to capture the non-Euclidean latent anatomy of human motions. We then incorporate the motion features from both Hyperbolic and Euclidean spaces, allowing us to precisely model the non-linear joint correlations while effectively sketching human poses. The proposed method empirically achieves state-of-the-art performance on the NTU RGB+D 60, NTURGB+D 120, and NW-UCLA datasets.

1 April 2026 in IEEE Transactions on Image Processing
Verified
Journal Article
Closed access
YJYingying Jiao
Yingying Jiao
HCHaipeng Chen
Haipeng Chen
YLYingda Lyu
Yingda Lyu
ZYZhaotin Yin
Zhaotin Yin
SWShuang Wu
Shuang Wu
ZLZhenguang Liu
Zhenguang Liu
Action recognition has long been a fundamental and compelling problem in the field of computer vision. However, one aspect that has been overlooked so far is that current action recognition approaches often produce an unfavourable multi-peaked distribution when identifying the action class of a given motion sequence, which is ambiguous and hard to learn for neural networks. Moreover, current methods heavily rely on neural networks to extract action features for differentiating actions, lacking theoretical constraints ensuring that action-specific features are selectively extracted and ambiguous features common to multiple actions are effectively reduced. These shortcomings culminate in inadequate action recognition accuracy. Motivated by this, in this paper we seek to tackle the problem from three aspects: (1) We try to eliminate ambiguity by enforcing a smooth single-peaked distribution instead of a multi-peaked one for action-class prediction. (2) We theoretically analyze the lower bound of the label prediction log-likelihood and derive a training objective, which focuses on the extraction of action-specific features and the reduction of ambiguous features. (3) We further advocate feeding the model with richer information, including positive information like body-part structures and negative information like masked inputs. Empirically, our approach sets the new state-of-the-art performance on five large-scale benchmarks. Our code is released at https://github.com/ActionR-Group/DPM to facilitate future research and is available in the supplementary material.

2 January 2026 in Sensors
Verified
Journal Article
Open access
PDF file
ZKZhejun Kuang
Zhejun Kuang
ZYZhaotin Yin
Zhaotin Yin
ZYZhaotin Yin
Zhaotin Yin
JZJian Zhao
Jian Zhao
LSLei Sun
Lei Sun
Action quality assessment automates the evaluation of human movement proficiency, which is vital for applications like sports training and rehabilitation, where objective feedback enhances patient outcomes. Action quality assessment processes motion capture data to generate quality scores for action execution. In rehabilitation exercises, joints typically work synergistically in functional groups. However, existing methods struggle to accurately model the collaborative relationships between joints. Fixed joint grouping is not flexible enough, while fully adaptive grouping lacks the guidance of prior knowledge. In this paper, based on rehabilitation theory in clinical medicine, we propose a dynamic, motion-aware grouping strategy. A two-stream architecture independently processes joint position and orientation information. Fused features are adaptively clustered into 6 functional groups by a joint motion energy-driven learnable mask generator, and intra-group temporal modeling and inter-group spatial projection are achieved through two-stage attention interaction. Our method achieves competitive results and obtains the best scores on most exercises of KIMORE, while remaining comparable on UI-PRMD. Experimental results using the KIMORE dataset show that the model outperforms current methods by reducing the mean absolute deviation by 26.5%. Ablation studies validate the necessity of dynamic grouping and the two-stream design. The core design principles of this study can be extended to fine-grained action-understanding tasks such as surgical operation assessment and motor skill quantification.