search

yzhq97 / transmomo.pytorch

This is the official PyTorch implementation of the CVPR 2020 paper "TransMoMo: Invariance-Driven Unsupervised Video Motion Retargeting".
https://yzhq97.github.io/transmomo/
392 stars 54 forks source link

the format of data #2

Closed ronghui19 closed 4 years ago

ronghui19 commented 4 years ago

i checked the format of data. I download mixamo data. When i decode npy format data, the shape is (15, 3, ...). The first dim is not 2. Why is that?

ronghui19 commented 4 years ago

The dataset i downloaded from google dont work correctly. I test many different npy file. All end up with shape (15, 3, something).

goodmattg commented 4 years ago

Mixamo dataset is the joint positions of 3D figures, so [15,3] is the [x,y,z] coordinates for 15 joints. TransMoMo uses the first 15 joints in the list of BODYPOSE_25

Walter0807 commented 4 years ago

@goodmattg Thanks for your explanation.

ronghui19 commented 4 years ago

Mixamo dataset is the joint positions of 3D figures, so [15,3] is the [x,y,z] coordinates for 15 joints. TransMoMo uses the first 15 joints in the list of BODYPOSE_25

thank you very much. One more question, for meanpose.npy, stdpose.npy, meanpose_with_view.npy, stdpose_with_view.npy, how am i supposed to calculate? I found out data.py in lib dir do the job for meanpose.npy and stdpose.npy.

what about meanpose_with_view.npy, stdpose_with_view.npy? what dataset can help me calculate it?