bug report and some notice

[gskrpb]

pkgs: batchgenerators 0.23

report bugs:

1. the same bug @CaiziLee mentioned in cannot reproduce the reported results #3 in dataset/acdc.py occured on dataset/chd.py
2. the similar bug above occured on dataset/hvsmr.py and dataset/mmwhs.py from batchgenerators.dataloading import MultiThreadedAugmenter from batchgenerators.transforms import Compose, RndTransform from batchgenerators.transforms import SpatialTransform, MirrorTransform from batchgenerators.transforms import GammaTransform, ConvertSegToOnehotTransform from batchgenerators.transforms import RandomCropTransform should be corrected as from batchgenerators.dataloading.multi_threaded_augmenter import MultiThreadedAugmenter from batchgenerators.transforms.abstract_transforms import Compose, RndTransform from batchgenerators.transforms.spatial_transforms import SpatialTransform, MirrorTransform from batchgenerators.transforms.color_transforms import GammaTransform from batchgenerators.transforms.utility_transforms import ConvertSegToOnehotTransform from batchgenerators.transforms.crop_and_pad_transforms import RandomCropTransform

notice: model = torch.nn.DataParallel(model, device_ids=args.multiple_device_id) in train_contrast.py reported error, which is due to parser.add_argument("--multiple_device_id", type=tuple, default=(0,1)) in myconfig.py but in fact, when it comes to multiple GPUs, maybe we should name the specific GPUs we request, like

os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" os.environ["CUDA_VISIBLE_DEVICES"] = "2, 3"

in train_contrast.py before the main function.

[dewenzeng]

@gskrpb Thank you very much for pointing out the bugs. These two issues have been addressed in the latest update.