IoU val zero

[xsk07]

Hello, I am having some troubles with the results of the evaluation step. What I'm trying to do is to run the model on the DeepGlob dataset in such a way that it ignores just the 'unknown' class, but what happens is that, if I'm using the 'default' configuration ignore_index: 6 and ignore_gt_labels: [6] then the model ignores both the classes 0 and 6 i.e. it returns 0 IoU for both of them. Following there is reported an example of history_epoch file showing the problem I'm referring to:

epoch                         0.000000
epoch.1                       1.000000
train_loss                    1.078230
train_acc                    49.376525
val_miou                      0.273724
val_acc                      58.940513
val_deformed_miou             0.274391
val_y_reverse_miou                 NaN
num_valid_samples           207.000000
train_edge_loss               0.652420
val_iou_class_0               0.000000
val_iou_deformed_class_0      0.000000
val_iou_class_1               0.672547
val_iou_deformed_class_1      0.667506
val_iou_class_2               0.181326
val_iou_deformed_class_2      0.176898
val_iou_class_3               0.420982
val_iou_deformed_class_3      0.426324
val_iou_class_4               0.406174
val_iou_deformed_class_4      0.414650
val_iou_class_5               0.235043
val_iou_deformed_class_5      0.235360
val_iou_class_6               0.000000
val_iou_deformed_class_6      0.000000

A similar behaviour happens also in the case when I'm using the configuration ignore_index: 0 and ignore_gt_labels: [0] (which I think it is the setting that I have to use in order to ignore just the 'unknown' class), more precisely in this case the "ignored" classes are the classes 0 and 1. How can I solve this problem and what I'm doing wrong? Thank you so much in advance :)