In deeprobust/image/defense/pgdtraining.pytroch.save(), there is a problem with the save path splicing, because the default value of save_dir does not have '/' at the end. So I use os.path.join to fix the path splicing process. Also, I have applied this approach to other torch.save().
In
deeprobust/image/defense/pgdtraining.pytroch.save(), there is a problem with the save path splicing, because the default value ofsave_dirdoes not have '/' at the end. So I useos.path.jointo fix the path splicing process. Also, I have applied this approach to other torch.save().