How can I inference RGBpose_conv3d??

[hotelbread]

Hello. I usually used mmaction2, but I moved to pyskl cause I only needed to use Rgbpose_conv3d model.

I appreciated your guide, how to train this model, I could have successed to train from my own dataset.
rgb_only
pose_only
and then merged these two

however, I met new problem when I try to inference by demo_skeleton.py
Im not sure whether it was good choice or not, but I didn't have any better options.

If you don't mind, can you guide for me? or do you have other guid for this?