yurujaja/dginstyle-segmodel

Downstream semantic segmentation evaluation of DGInStyle.

dginstyle-segmodel's Issues

Undestanding the training configs

Hi,

Thanks for making all the components of the paper (including those related to segmentation training) available!

I see that the config folder has many config files present, but it is unclear which one corresponds to what setting, or how to understand the file names. For example, it seems like uda_gengtaCAug_to_cityscapes_512x512.py is the default config for exp-id 60. In this case, what do the following mean: uda, gengtaCAug, to_cityscapes. Understanding this would help me better design my own experiments.

DGDataset indexing for training

Could you please explain how the datasets are getting samples?

Suppose I am trying to run exp-id 60 with the GTA and DG datasets. Here I see that initially the source dataset is sampled and then source2. Is that correct?

Can you also confirm the expected sizes of these two datasets?

The generated Cityscapes dataset

Hello! This is a nice work. Could you plan to release the generated Cityscapes dataset?