Small question about training.py in johnson_alahi_li_2016. The original authors have published hyperparameters for the different styles, which we also use in the replication. Among other things, the hyperparameter num_batches have to be adjusted. Doesn't image_loader in main(args) therefore have to be created later with the adapted hyperparameters? Or am I missing something there?
Small question about training.py in
johnson_alahi_li_2016
. The original authors have published hyperparameters for the different styles, which we also use in the replication. Among other things, the hyperparameternum_batches
have to be adjusted. Doesn'timage_loader
inmain(args)
therefore have to be created later with the adapted hyperparameters? Or am I missing something there?Proposal change: