search

c-he / NeMF

[NeurIPS 2022] Official implementation of "NeMF: Neural Motion Fields for Kinematic Animation"
MIT License
156 stars 9 forks source link

How to train the generative NeMF on dog mocap dataset? #21

Closed jonggul2 closed 5 months ago

jonggul2 commented 5 months ago

Hello, I would like to train a Generative NeMF model on dog mocap data, but only the code for training a Single-motion NeMF model is available. Could you provide the code for this, and also let me know if there is a pre-trained model available?

c-he commented 5 months ago

Hi, thanks for your interest in our project. Sorry that we don't have a pre-trained model on dog mocap data as we only use them for validation. You need to mofify our codebase to train this model, and the logic would be:

  1. Modify animal.py to process all the dog mocap data and split them into the training/validation/testing sets (currently we process each sequence one by one and store them separately).
  2. Modify train.py to load the processed Animal Dataset rather than AMASS, and write a corresponding config file to properly set the hyperparaemters.
  3. Modify train_gmp.py to train it for the Animal Dataset rather than AMASS, a new config file would be required as well.

There might be other issues I overlooked, but the basic logic is here. If you run into any issues, welcome to point them out and I can help to resolve them.

jonggul2 commented 5 months ago

Thank you.