How can I train a model tracking the image-level AUROC (classification)

[mjack3]

Let's suppose the following code

anomalib train --config configs/model/csflow.yaml --data --configs/data/mvtec.yaml

I have set in the model config file:

model:
  class_path: anomalib.models.Csflow
  init_args:
    cross_conv_hidden_channels: 1024
    n_coupling_blocks: 4
    clamp: 3
    num_channels: 3

metrics:
  image:
    - AUROC

trainer:
  max_epochs: 20
  callbacks:
    - class_path: lightning.pytorch.callbacks.EarlyStopping
      init_args:
        patience: 100
        monitor: image_AUROC
        mode: max

and the following in the data config file

task: classification

But the pixel auroc is always tracked during training. Am I missing something?