I try to recognize if any animal is present (non_blank) or not (blank) on the camera trap images:
I tried to change the metric from accuracy
to auc_roc_score
, but I’m getting the following error:
The size of tensor a (121) must match the size of tensor b (2) at non-singleton dimension 1
My questions are:
- As far as I understand, my model is expecting a multiclass problem, but it’s getting binary metric
auc_roc_score
. What should I change so that I’ll be able to useauc_roc_score
? -
Do you think that
auc_roc_score
is not the best one for this purpose? If yes - please share your thoughts.
My folder structure is as follows:
I load the data using ImageDataBunch.from_folder
:
np.random.seed(42)
size=144
data = ImageDataBunch.from_folder(p, seed=123, padding_mode='zeros', size=size, num_workers=4, resize_method=ResizeMethod.SQUISH,
ds_tfms=get_transforms(max_rotate=2.0, max_zoom=0.9, max_lighting=0.2, max_warp=None)
).normalize(imagenet_stats)