search

marcoaversa / diffinfinite

DiffInfinite Official Code
MIT License
23 stars 3 forks source link

why always dropping conditionnal probability in test set ? #7

Open joihn opened 3 months ago

joihn commented 3 months ago

Hello, thanks for this repo.

    test_set = DatasetLung(data_path=data_path, data_dict=test_dict, 
                           subclasses=subclasses, cond_drop_prob=1.,
                           transform=transform)

https://github.com/marcoaversa/diffinfinite/blob/4d2309606e5ce6190828b96f436cfa962ec5d1ec/dataset.py#L274

why in the test set, cond_drop_prob=1. ? This will always drop the conditionnal (=the mask ) during evaluation. In my opinion, to have a test as clos eas possible to final task (imge-mask pair generation) one should keep the conditionnal (on should set cond_drop_prob=0.), what do you think ?