.png){kind=icon}
Return
Human motion prediction using knowledge distillation with incremental learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s10044-026-01703-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Human motion prediction has traditionally concentrated on spatial-temporal modeling and network architecture development, with relatively limited exploration of the transmission and processing of information flow. Incremental learning, which has emerged as a promising direction, aims to reduce noise interference and capture deeper features through finer-grained inputs. The architecture of Graph Convolutional Networks (GCNs) is particularly well-suited for this task, as it effectively captures the spatial-temporal characteristics of human motion, opening new avenues for research. GCNs excel at modeling the complex spatial-temporal relationships inherent in human motion data, making them ideally suited for incorporating advanced learning techniques. In this paper, we introduce a novel spatial-temporal GCN that incorporates incremental learning enhanced by knowledge distillation techniques. This combination allows the model to reformulate motion data, improving feature fusion and effectively reducing noise interference. Additionally, an innovative input smoothing technique is introduced to preprocess motion data, addressing the inherent variability and noise in raw data. Such preprocessing ensures more reliable inputs, thereby enhancing the robustness and accuracy of predictions. Evaluated using the Mean Per Joint Position Error (MPJPE), the model demonstrates superior accuracy compared to state-of-the-art methods.
Keywords:
Human motion prediction
Incremental Learning
Knowledge distillation
Graph Convolutional Network
Journal
IF:
2
Papers:
1.9K
Citations:
1.9K
Organization
Cited Papers
ShareSave
ShareSaveKD-Former: Kinematic and dynamic coupled transformer network for 3D human motion prediction.png)
PATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
ShareSaveShareSave
View more