search

bastianwandt / RepNet

This is the original RepNet implementation
86 stars 17 forks source link

data process in Repnet #12

Open laihuaijing opened 2 years ago

laihuaijing commented 2 years ago

Thanks for opening source, I have downloaded the human3.6m dataset from offical website, but I am not quite clear about aligning the 3D pose to a template as described in the paper, so I wonder if you can share the scripts for preprocessing the dataset?

My approach is 

d, Z1, tform1 = procrustes(self.mean_shape, pose_3d[[0,3,10,13],:])
r=tform1['rotation']
b=tform1['scale']
t=tform1['translation']
pose_3d=np.matmul(pose_3d,r)*b+t

and I calculate MPJPE directly using the output of the generator without post-processing, is it right?

Looking forward to your reply.