Mean Shift Rejection: Training Deep Neural Networks Without Minibatch Statistics or Normalization

no code implementations • 29 Nov 2019 • Brendan Ruff, Taylor Beck, Joscha Bach

This new technique allows direct training of the test graph so that training and test models are identical.

Data Augmentation Unity