Researchers have introduced HSTGFormer, a novel graph-enhanced Transformer framework designed for monocular 3D human pose estimation. This model reformulates spatial-temporal reasoning by employing localized, coupled graph aggregation over joint-time nodes. It utilizes a Hyper Spatial-Temporal Graph (HSTG) to decompose reasoning into local receptive fields and an Adaptive Dual-Scale Temporal Graph (ADSTG) to capture joint-specific temporal dependencies. Experiments on Human3.6M and MPI-INF-3DHP datasets demonstrate that HSTGFormer achieves high accuracy with notable computational efficiency. AI
IMPACT This new model offers improved accuracy and computational efficiency for 3D human pose estimation tasks.
RANK_REASON The cluster contains a research paper detailing a new model for 3D human pose estimation. [lever_c_demoted from research: ic=1 ai=1.0]
- Adaptive Dual-Scale Temporal Graph
- arXiv
- HSTGFormer
- Human3.6M: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments
- Hyper Spatial-Temporal Graph
- MPI-INF-3DHP
- Transformer++
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
