Awesome
3DV: 3D Dynamic Voxel for Action recognition in depth video.
Paper pdf
To be represented in CVPR2020
Usege:
Our model consists of 2 parts. The first part is about 3DV point data generation and apprearance point data preparation. For example, "3DV_construction\python\ntu120_3dv_pre.py" is used to generate the 3DV point data with multiple motion features. The output 3DV points data form is N*(x,y,z,m_g,m_1,...,m_T) "3DV_construction\python\RawDepthPoints_appearance_pre.py" is used to sample multi-segment temporal frames with its raw points into the form of N*(x,y,z).
The second part is about multi-stream PointNet++. The generated point data in part 1 can be fed to Network.
Some related data: