Question about SMPL tracking

[tongxueqing]

Thanks for you wonderful open-source work! How do you do SMPL tracking to initialize? Why the initial shape parameter is not learnable but the pose per frame is learnable?

[jby1993]

I use VideoAvatar or easymocap to initialize, you can use others too. The canonical shape is represented by an SDF and can be refined, the SMPL shape is only used to initialize the SDF.