Normalize test set?

(Vincenzo) #1

Hi, this is my first question, although I have been a reader of the forum since some time (thank you all).

I am a recent user of Fastai. I am able to train and do some inference on images. What I am not yet able (or sure to be able) is to normalize a test set. The code I use is:

path= …
test = ImageList.from_folder(os.path.join(path, ‘unknown/’))
learner = load_learner(path, file=‘model.pkl’, test=test)
preds, y = learner.get_preds(ds_type=DatasetType.Test)

Thanks.

0 Likes

(Zachary Mueller) #2

When we call load_learner, we do an add_test(test), and when we do this we load our model with the same data statistics as our training set originally was, so this is applied :slight_smile:

(see the very bottom of the basic_train file (https://github.com/fastai/fastai/blob/master/fastai/basic_train.py))

1 Like

(Vincenzo) #3

Thanks, I supposed something like this but I did not know how to prove it.

0 Likes