search

DeepMotionEditing / deep-motion-editing

An end-to-end library for editing and rendering motion of 3D characters with deep learning [SIGGRAPH 2020]
BSD 2-Clause "Simplified" License
1.57k stars 257 forks source link

Request for Characters with "_m" #139

Closed loveyourdaddy closed 3 years ago

loveyourdaddy commented 3 years ago

Hi, I'm a graduate student researching motion retargeting. I am excited about your work and paper. I ran your code and tested the network on new motion data with Mixamo. I noticed characters with "_m" are not the same as them in Mixamo. Could you provide me these 4 characters for research purposes?

Thanks in advance.

PeizhuoLi commented 3 years ago

Thanks for your interest in our project. As we described in README and paper, we manually split three joints in group A. The "_m" suffix is used to denote the character that has been processed. You may use datasets/split_joint.py to get the character processed. For more details please check README.

loveyourdaddy commented 3 years ago

Thanks for the quick answer! It helps me for testing the network model.

I have one further question. Have you tested on validation data which is un-seen motion in the training process for cross-structural retargeting? The pre-trained model and the model I trained seem not retargeting motion well on motion data not included in the training set.

Again, thanks for the brilliant research.

PeizhuoLi commented 3 years ago

You are welcome : )

In our experiment we split the dataset into two parts (training and test) following NKN. The model worked well on the test set.

If it's possible you can provide some example on the failure cases so I can understand better about the details.

loveyourdaddy commented 3 years ago

There was one mistake in me using the old npy file. It is working well now. Sorry for the confusion and thank you for your kind answer.