Register a semantic segmentation datasets

[trongan93]

I'm custom a dataset for a DeepLab V3 model (following the DeepLab implementation under projects).
My data is a binary dataset (label 1 for object and 0 for background), and my register dataset code is below.

def registry_dataset_semantic_segmentation():    
    name = "custom_dataset_train"
    # 1. register a function which returns dicts
    DatasetCatalog.register(name, lambda: load_sem_seg('/mnt/d/train/PS-RGB_tiled_mask', '/mnt/d/train/PS-RGB_tiled', gt_ext='png', image_ext='png'))

    # 2. Optionally, add metadata about this dataset,
    MetadataCatalog.get(name).set(thing_classes=["aircraft"])
    
    name_test = "custom_dataset_test"
    # 1. register a function which returns dicts
    DatasetCatalog.register(name_test, lambda: load_sem_seg('/mnt/d/test/trongan_PS-RGB_tiled_mask', '/mnt/d/test/PS-RGB_tiled', gt_ext='png', image_ext='png'))
    # 2. Optionally, add metadata about this dataset,
    MetadataCatalog.get(name_test).set(thing_classes=["aircraft"], evaluator_type=["sem_seg"],stuff_classes=[""], ignore_label=[0])

My training is passed. However, the accuracy and loss values are weird. Actually, both values are incorrect.
I got an error in the visualization process.

[05/07 07:00:01 d2.evaluation.evaluator]: Start inference on 245 batches
[05/07 07:00:21 d2.evaluation.evaluator]: Inference done 11/245. 0.0425 s / iter. ETA=0:00:13
[05/07 07:00:26 d2.evaluation.evaluator]: Inference done 96/245. 0.0440 s / iter. ETA=0:00:08
[05/07 07:00:31 d2.evaluation.evaluator]: Inference done 182/245. 0.0440 s / iter. ETA=0:00:03
[05/07 07:00:35 d2.evaluation.evaluator]: Total inference time: 0:00:14.732784 (0.061387 s / iter per device, on 7 devices)
[05/07 07:00:35 d2.evaluation.evaluator]: Total inference pure compute time: 0:00:10 (0.044370 s / iter per device, on 7 devices)
trongan93 - test - acc: [nan] , iou: [nan]
/home/eeaiserver/viplab_projects/detectron2/detectron2/evaluation/sem_seg_evaluation.py:135: RuntimeWarning: invalid value encountered in true_divide
  class_weights = pos_gt / np.sum(pos_gt)
/home/eeaiserver/viplab_projects/detectron2/detectron2/evaluation/sem_seg_evaluation.py:142: RuntimeWarning: invalid value encountered in double_scalars
  macc = np.sum(acc[acc_valid]) / np.sum(acc_valid)
/home/eeaiserver/viplab_projects/detectron2/detectron2/evaluation/sem_seg_evaluation.py:143: RuntimeWarning: invalid value encountered in double_scalars
  miou = np.sum(iou[acc_valid]) / np.sum(iou_valid)
/home/eeaiserver/viplab_projects/detectron2/detectron2/evaluation/sem_seg_evaluation.py:145: RuntimeWarning: invalid value encountered in double_scalars
  pacc = np.sum(tp) / np.sum(pos_gt)
[05/07 07:00:35 d2.evaluation.sem_seg_evaluation]: OrderedDict([('sem_seg', {'mIoU': nan, 'fwIoU': 0.0, 'IoU-': nan, 'mACC': nan, 'pACC': nan, 'ACC-': nan})])
[05/07 07:00:35 d2.engine.defaults]: Evaluation results for custom_dataset_test in csv format:
[05/07 07:00:35 d2.evaluation.testing]: copypaste: Task: sem_seg
[05/07 07:00:35 d2.evaluation.testing]: copypaste: mIoU,fwIoU,mACC,pACC
[05/07 07:00:35 d2.evaluation.testing]: copypaste: nan,0.0000,nan,nan

Are someone else has worked on a custom dataset on semantic segmentation? I guess my registration dataset has some errors, but I not sure how to fix them.

Thank you for your attention.