Multi-label show_results on test data

Part 1 (2019)
1

Hi I’m working on Lesson3 multi-label prediction but struggling to make show_results work test data. I know this wasn’t the part of the original Lesson 3 notebook but I just want to visualize the predictions. The problem is described first and then the source code will follow.

show_results on the validation works as expected:

show_validation
show_validation.png1125×1245 1.02 MB

However, it behaves weird on the test data:
learn.show_results(ds_type=DatasetType.Test)

show_test
show_test.jpg1125×1245 692 KB 

src = (ImageList.from_csv(path, 'data.csv', folder='train')
   .split_by_rand_pct(0.2)
   .label_from_df(label_delim=' ')
   .add_test_folder())
data = (src.transform(get_transforms(), size=(128,128))
    .databunch().normalize(imagenet_stats))
data.test_ds

LabelList (25 items)
x: ImageList
Image (3, 128, 128),Image (3, 128, 128),Image (3, 128, 128),Image (3, 128, 128),Image (3, 128, 128)
y: EmptyLabelList
,
Path: /media/kra/TTera/dataset/spongebob_multi

acc_02 = partial(accuracy_thresh, thresh=0.2)
f_score = partial(fbeta, thresh=0.2)
learn = cnn_learner(data, arch, metrics=[acc_02, f_score])
# and do some training...
1 Like