neodroidvision.utilities.deep_augmentation.deep_batch_augmentation.gauss_sample_deep_aug

neodroidvision.utilities.deep_augmentation.deep_batch_augmentation.gauss_sample_deep_aug()[source]

Use category balanced batches, ensuring at least 3-5 instance of a category. Then per category in batch, find local manifold by finding the covariance matrix at

representation level

and sample a multi var gauss for new samples for further forward passes