[Python] How to get transformation matrix from Kornia Augmentation pipeline?

I have a created a Kornia augmentation pipeline for training that looks like:

self.train_augments = AugmentationSequential(
K. VideoSequential(
Normalize255(),
K.ColorJitter(0.3, 0.3, 0.3, 0.1, p=0.8),
K.Resize(resize_to),
K.RandomRotation(degrees=12.0, p=0.5, same_on_batch=True),
K.CenterCrop(resize_to),
K.RandomHorizontalFlip(p=0.5, same_on_batch=True),
K.Normalize(mean, std),
same_on_frame=True

),
data_keys=["input", "keypoints"],
)


Normalize255 is a function that i have implemented myself because of pickling errors. it subclasses nn.Module and basically normalizes the image by a factor of 255.0

My question is: in this pipeline, since RandomHorizontalFlip happens with a probability of 0.5, how do i know in which cases the flip augmentation has been applied? i ask because, if flip has been applied, i will need to update my keypoints using flip-indices.

Continue reading...