search

Jeff-sjtu / HybrIK

Official code of "HybrIK: A Hybrid Analytical-Neural Inverse Kinematics Solution for 3D Human Pose and Shape Estimation", CVPR 2021
MIT License
1.21k stars 147 forks source link

Bug encountered with flip_test #137

Open Bingrong89 opened 1 year ago

Bingrong89 commented 1 year ago

image Hello, under hybrik/models/simple3dposeSMPLwithCam.py, image

Can you confirm if at line 272, whether the intended input to self.decsigma is flip_x0 or should it actually be flip_xc? flip_x0 is shape (batchsize,512) while flip_xc is (batchsize,1024), while the layer self.decsigma is a fully connected layer expecting input with feature length 1024.

Bingrong89 commented 1 year ago

image Also this, anyone knows what this is supposed to be?

MrMiilk commented 1 year ago

The same issue, could you let me know if you solved it?

Fouksi commented 1 year ago

I have the same issue. Did you guys manage to fix it?

daixiangzi commented 1 year ago

image Hello, under hybrik/models/simple3dposeSMPLwithCam.py, image

Can you confirm if at line 272, whether the intended input to self.decsigma is flip_x0 or should it actually be flip_xc? flip_x0 is shape (batchsize,512) while flip_xc is (batchsize,1024), while the layer self.decsigma is a fully connected layer expecting input with feature length 1024.

I should be self.decsigma(filp_xc)