The class DenseCorrespondenceTraining (in training.py) manages the training process. It takes in a config specifying parameters. See config/dense_correspondence/training/training.yaml for an example. You just need to call run() on the DenseCorrespondenceTraining object to run the training. Plotting is done using Visdom, navigate to localhost:8097 in the browser and choose the appropriate env from the dropdown menu for your training run. Note that the env will be the same as the name of the logging directory for that training run.
See the notebook training.ipynb for an example of how to use this class.