search

sail-sg / mvp

NeurIPS-2021: Direct Multi-view Multi-person 3D Human Pose Estimation
Apache License 2.0
325 stars 34 forks source link

Question about '3D joint position y' in Projective Attention #30

Open Billccx opened 10 months ago

Billccx commented 10 months ago

Hello!

I'm wondering what y refers to in the Projective Attention module.

Is it the ground truth of the 3D keypoints? If so, how do we handle it when the model is inferencing?

I'm looking forward to your reply and thanks in advance ~

image

image

twangnh commented 10 months ago

hi it means the 3d joint location prediction of the current decoder layer

Billccx commented 10 months ago

hi it means the 3d joint location prediction of the current decoder layer

Thanks for your quick reply !

I have checked the source code. It seems that reference_points is y ?

Does this mean that y is actually the reference_points output by the previous decoder layer?

https://github.com/sail-sg/mvp/blob/8b2ccc576a450841a5b344597cec26e5ac77eaf7/lib/models/mvp_decoder.py#L314-L345

I am not sure if my understanding is correct. If it is not, could you please explain it in detail? Thank you.