The opt eval flag is currently not giving the desired behaviour. The flag gets reset for all dataset the same, resulting after the first evaluation in a disabling of the augmentation pipeline. This should make sure it behaves a expected, a more proper solution would be copy over all the specific attributes for the datasets classes.
The opt eval flag is currently not giving the desired behaviour. The flag gets reset for all dataset the same, resulting after the first evaluation in a disabling of the augmentation pipeline. This should make sure it behaves a expected, a more proper solution would be copy over all the specific attributes for the datasets classes.