I am training a custom dataset (RarePlane) with DeepLab V3+ using Detectron2 (open source are wroten base on Pytorch).

The custom dataset is fixed with an image size is 512x512. When I trained with 100000 interactions, I got the mIoU values (bellow).

[05/10 06:13:49] d2.evaluation.sem_seg_evaluation INFO: OrderedDict([('sem_seg', {'mIoU': 48.263697089435894, 'fwIoU': 93.17537826963293, 'IoU-a': nan, 'IoU-i': 0.0, 'IoU-r': nan, 'IoU-c': nan, 'IoU-f': nan, 'IoU-t': nan, 'mACC': 50.0, 'pACC': 96.52739417887179, 'ACC-a': nan, 'ACC-i': 0.0, 'ACC-r': nan, 'ACC-c': nan, 'ACC-f': nan, 'ACC-t': nan})]) [05/10 06:13:49] d2.engine.defaults INFO: Evaluation results for custom_dataset_test in csv format: [05/10 06:13:49] d2.evaluation.testing INFO: copypaste: Task: sem_seg [05/10 06:13:49] d2.evaluation.testing INFO: copypaste: mIoU,fwIoU,mACC,pACC [05/10 06:13:49] d2.evaluation.testing INFO: copypaste: 48.2637,93.1754,50.0000,96.5274

I'm looking for a solution to configuring the DeepLab code with detectron2 and how to increase mIoU values.

Thanks.