VARPose: Flexible 2D Pose Densification via Visual Autoregressive Modeling for Enhanced 3D Lifting
VARPose:基于视觉自回归建模的灵活2D姿态密集化以提升3D姿态提升性能
机构 * School of Artificial Intelligence, Sun Yat-sen University(中山大学人工智能学院) ; The Technology Innovation Center for Collaborative Applications of Natural Resources Data in GBA, MNR(自然资源部粤港澳大湾区自然资源数据协同应用技术创新中心)
AI总结 VARPose基于VAR建模,通过GPT分词器和UniSkelar自回归模型实现2D姿态灵活密集化,在3D姿态估计等下游任务上性能优于现有方法且可泛化到新粒度。
Comments ACM MM 2026