Test-Time-Augmentation(TTA)

[chuxiang93]

Hi,
I wonder if you can add some tools for test-time-augmentation, especially in semantic segmentation tasks.
Thank you

[Nic-Ma]

Hi @ShuangGou ,

Thanks for your feature request.
TTA is very useful for model evaluation. I will try to develop several other EnsembleEvaluators(average, vote, etc.) besides existing SupervisedEvaluator under engines, and then develop some tutorial to show the usage.

Thanks.

[tvercaut]

@taigw Feel free to chime in, it would be great to have an implementation of our paper in MONAI:
Wang, Guotai, Wenqi Li, Michael Aertsen, Jan Deprest, Sébastien Ourselin, and Tom Vercauteren. "Aleatoric uncertainty estimation with test-time augmentation for medical image segmentation with convolutional neural networks." Neurocomputing 338 (2019): 34-45. https://doi.org/10.1016/j.neucom.2019.01.103

[wyli]

this partly depends on #365