Question about the part probabilities

[Bingoang]

@bharat-b7
Hi, Bharat~
I really interesting about your paper and code, but I have a question about the part probabilities:
For each query point p, you calculate a probability parts_softmax (shape: B×14×N) to judge which part of the SMPL the query point p belongs to. (The related code is in the line 124: https://github.com/bharat-b7/IPNet/blob/master/models/local_model_body_full.py)
You divide the SMPL body into 14 parts. Why is the shape of probability parts_softmax not B×(14+2)×N? What if the query point p does not belong to any part of the SMPL body (e.g. point p is inside or outside the surface)?

I cannot figured it out, and I really hope you can answer me. Thanks in advance~

[bharat-b7]

We predict the label of the closest SMPL part. This independent to weather the point is inside or outside, which is handled by a separate occupancy prediction.

[Bingoang]

@bharat-b7
Oh I see~ Thank you so much for your quick reply!