bone2skeleton conversion

[snitchjinx]

Hello, thanks for sharing this nice work!

I have a question regarding the bone to skeleton conversion function in ./models/model.py It seems the joints are shifted from the pelvis joint by offsets of bone lengths, but why is there no horizontal offsets for RKnee, RAnkle, LKnee, and LAnkle. I was expecting something like:

skel_in[:, 2, 0] = -unnorm_bones[:, 0, 0] skel_in[:, 5, 0] = unnorm_bones[:, 0, 0] skel_in[:, 3, 0] = -unnorm_bones[:, 0, 0] skel_in[:, 6, 0] = unnorm_bones[:, 0, 0]

Thanks in advance for consideration!

[Shimingyi]

Hello! A good question!

We use this function to convert the bone length to T-skeleton, in here we haven't apply the horizontal offsets on Knees and Ankles because we found this kind of setting will be more similar to various character model, so it's better for skinning and other processes.

But because we only apply positional loss directly in our training, so the model can also converage if you use different setting.

[snitchjinx]

Thanks!

[longbowzhang]

Hi @Shimingyi I am also curious about the bones2skel function in model.py. I guess this function constructs the skeleton in terms of relative positions based on the following two oberservations.

• skel_in[:, 2, 1] = -unnorm_bones[:, 0, 1] means the position of RKnee is unnorm_bones[:, 0, 1] below (i.e., negative y-axis) its parent joint RHip, and there is no horizontal offset w.r.t RHip.
• as shown in the function of _transformsglobal in model_zoo.py, the positions (i.e., the constructed skeleton) is first used to build locals.

But I would like to double check with you about bones2skel whether it expresses the t-pose skeleton in terms of relative positions or absolute positions.

Many thanks.

[Shimingyi]

Hi@longbowzhang ,

You are right, it's not the absolute position in a 3d space. Each value is the relative position compared to its parent joint. This code can help you to undersantd it better: get_skel

Best, Mingyi