4DHands: Reconstructing Interactive Hands in 4D with Transformers

Dixuan Lin^1,2, Yuxiang Zhang², Mengcheng Li², Yebin Liu²,
Wei Jing³, Qi Yan³, Qianying Wang^3*, Hongwen Zhang^1*

¹Beijing Normal University, ²Tsinghua University, ³ Lenovo

Code

4DHands can robustly recover interactive hand meshes and their relative movement from monocular inputs.

Results - Monocular Video Input

4DHands outperforms other interactive hand reconstruction methods in real-world videos.

Results - Single Image Input

4DHands has robust performance in complex cases.

Results - Complex Interactions

4DHands has robust performance in complex cases.

Results - Multi-Cameras

4DHands can handle highly occluded cases when applied to multi-view tasks.