.png){kind=icon}
.png){kind=icon}
Return
Dual-stream spatio-temporal GCN-transformer network for 3D human pose estimation
.png){kind=icon}
PRE
AI
D
X
L
X
X
DOI：10.1016/j.displa.2026.103429.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
3D human pose estimation is a classic and important research direction in the field of computer vision. In recent years, Transformer-based methods have made significant progress in lifting 2D to 3D human pose estimation. However, these methods primarily focus on modeling global temporal and spatial relationships, neglecting local skeletal relationships and the information interaction between different channels. Therefore, we have proposed a novel method-the Dual-stream Spatio-temporal GCN-Transformer Network (MixTGFormer). This method models the spatial and temporal relationships of human skeletons simultaneously through two parallel channels, achieving an effective fusion of global and local features. The core of MixTGFormer is composed of stacked Mixformers. Specifically, the Mixformer includes the Mixformer Block and the Squeeze-and-Excitation Layer (SE Layer). It first extracts and fuses various information about human skeletons through two parallel Mixformer Blocks with different modes. Then, it further supplements the fused information through the SE Layer. The Mixformer Block integrates Graph Convolutional Networks (GCN) into the Transformer, enhancing both local and global information utilization. Furthermore, in order to capture the spatial and temporal relationships of the skeleton at the same time, we designed two forms of Mixformer Block: Spatial Mixformer Block and Temporal Mixformer Block. The Spatial Mixformer Block treats each joint as a token to model the spatial relationship within a single frame; on the contrary, the Temporal Mixformer Block treats each frame as a token to model the temporal dynamics between consecutive frames. We extensively evaluated our model on two benchmark datasets (Human3.6 M and MPI-INF-3DHP). The experimental results showed that, compared to other methods, our MixTGFormer achieved state-of-the-art results(As of March 2025), with P1 errors of 37.6 mm and 16.5 mm on these datasets, respectively. The code is available at https://github.com/FDGHuTutu/MixTGFormer/tree/master .
Keywords:
3D human pose estimation
Transformer
Graph convolution
Journal
IF:
3.4
Papers:
2.1K
Citations:
3.2K
Organization
Cited Papers