osc.data.utils

Helper functions to bridge tensorflow image data to pytorch.

Functions:

• Image normalization back and forth

• Augmentation

  • Augmentations with ‘’stateless’’ random seeds

  • Fixed augmentations

  • Double augmentation for contrastive training

Functions

augment_center_crop	Deterministic center crop and processing as in augment_train()
augment_train	Augment image for train/val.
augment_twice	Augment image twice with two different functions.
img_hwc_to_chw	Image channels [..., H, W, C] -> [..., C, H, W]
normalize_tf	Normalize tf.float32 image with [..., H, W, C] channel order.
unnormalize_pt	Un-normalize torch.float32 image with [..., C, H, W] channel order.
wrap_with_seed	Wrap stateless random functions by providing a new seed to each invocation.

Classes

AugmentFnStateless

Function type signature for TF stateless random.