This is a fork of the original repo from https://github.com/ezhan94/calibratable-style-consistency.

Code is written in Python 3.7.4 and PyTorch v.1.0.1.

Train models with:

$ python run_single.py -d <device id> --config_dir <config folder name>

Not specifying a device will use CPU by default. See JSONs in configs\ to see examples of config files.

$ python run_single.py --config_dir test --test_code should run without errors.

$ python scripts/check_dynamics_loss.py -f <config folder name> will compute and visualize the dynamics model error, where applicable.

$ python scripts/compute_stylecon_ctvae.py -f <config folder name> will compute the style-consistency.

$ python scripts/visualize_samples_ctvae.py -f <config folder name> will sample and save trajectories for each label class.