trained model and configs

[felix-ky]

Thank you for updating the code,

could you please update "amass_softplus.yaml" and the optimal model?

best wishes

[garvita-tiwari]

We are running some tests and will update the model and config asap.
Sorry for the delay

[felix-ky]

Thank you so much.

Do denoise results of humor in Table 1 computed by adding noise to rotation and then converted to 3Djoints as input to humor's TestOpt? as I found humor only performs denoising of 3Djoints.

best wishes

[garvita-tiwari]

Thank you so much.

Do denoise results of humor in Table 1 computed by adding noise to rotation and then converted to 3Djoints as input to humor's TestOpt? as I found humor only performs denoising of 3Djoints.

best wishes

Hi, we use the loss function described in eq 7 and 8 of the paper(section 4.2), we use both joint and SMPL vertices in our denoising.

[felix-ky]

Thank you so much

to confirm, denoising using humor in table 1 just change posndf prior to humor prior in eq 7 ?
could you update the code using humor to denoise?

best wishes