neodroidvision.utilities.deep_augmentation.deep_batch_augmentation.non_linear_interpolate_pairs_deep_aug

neodroidvision.utilities.deep_augmentation.deep_batch_augmentation.non_linear_interpolate_pairs_deep_aug()

Find a pair and use a learned interpolator to interpolate between points.

for the learned interpolation regularize the representation by adding a small amount of noise to the representation abd impose penalty on l2 norms, and so on to avoid degenaracy/mode collapse. Same a avoid overfitting a generative model.

Look at variational autoencoders for this.

OTHER NOTES: Ties to generative diffusion models?

Look up LINDA: learning to interpolate for data augmentation. and SSMBA: state space model for data augmentation. and Mixup: non-local data augmentation.

Domain specific: combine multiple Facial features from different people. set of eye from person A and mouth from person B.

Returns

Return type