Currently, `BILINEAR` (default) and `NEAREST` are supported.
0.8.2
One can now do `ds = seqs_to_tf_dataset(images, [], [])` to get a Dataset which returns `(image, [], [])` on each iteration. This is more convenient than creating dummy `bboxes` and `labels` when working with image data only.
0.8.1
Fix getting random transform steps in `TFCombineAffine`.
0.8.0
Add support for randomly selecting only a subset of transform steps to use, with new parameters `num_selected_transforms` and `selected_probabilites` introduced to `CombineAffine` and `TFCombineAffine`.
0.7.7
Change the default probability of the random flip to 0.5