Using `kornia` with V2 in a generic way

[kshitijpatil09 (Kshitij Patil)]

You can use transforms.RandomApply for the transforms that aren’t inherently “random” like so:

color_jitter = K.augmentation.ColorJitter(0.8*s, 0.8*s, 0.8*s, 0.2*s)
rnd_color_jitter = transforms.RandomApply([color_jitter], p=0.8)

Wrapping individual transform would be needed when you’ve different behavior for different types, like images would be resized using Bilinear interpolation and masks using nearest neighbour. If the set of trasforms are only meant for single fastai type, wrapping them in nn.Sequential and then in some fastai wrapper seems a better idea.

I think if we create separate wrappers for train_tfms and test_tfms like:

class TrainTfms(Transform):
    order=10
    split_idx=0
    def __init__(self,tfm):
        self.tfm=tfm
    def encodes(self,x:TensorImage): return self.tfm(x)
    def __repr__(self): <listing all the transforms>

class ValidTfms(Transform):
    order=10
    split_idx=1
    def __init__(self,tfm):
        self.tfm=tfm
    def encodes(self,x:TensorImage): return self.tfm(x)
    def __repr__(self): <listing all the transforms>

it would be possible to include them in batch_tfms as well. Override __repr__ method to list all the underlying transform so that it won’t just appear like KorniaWrapper

I had already stumbled upon this problem and had asked for some way to pass these parameters through constructor.

Fastai v2 vision

Need some way to define order and split_idx through constructor. I’m trying to use kornia augmentations as batch transforms and these are the only arguments I need to modify. The only way I could think of is subclassing the Transform and wrapping each and every augmentation with that class, just the way it’s done with Albumentations in pets tutorial Also, to clarify, Transform with split_idx=0 will be applied on train-set, split_idx=1 for valid/test and split_idx=None for both, am I right ?

@rsomani95

still can’t figure out what’s causing only the last of the transforms to be called in the example posted on the very top of this post.

this might be related to order of transform, try assigning different order for each transform