RandomPerspective and RandomAffine fail when used wih pytorch-lightning precision=16

Describe the bug

HI, I am using the following transform: train_transforms = torch.nn.Sequential( K.augmentation.RandomHorizontalFlip(), K.augmentation.RandomPerspective(distortion_scale=0.02), K.augmentation.RandomAffine(degrees=(-5.0, 5.0), translate=(0.02, 0.02), scale=(0.9, 1.1), shear=(-0.02, 0.02), resample=‘bilinear’), )

pl version: 1.5.4 kornia version: 0.6,1 pytorch version: 1.9.1+cu111

The error is: RuntimeError: expected scalar type float but found c10::Half

Reproduction steps

1.Create a simple PyTorch-lightning model
2.set precision=16 in the trainer parameters
3.
...

Expected behavior

Support 16bit precision.

Environment

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

• PyTorch Version (e.g., 1.0):
• OS (e.g., Linux):
• How you installed PyTorch (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information:

### Additional context

_No response_

[Borda]

I think that the half-precision on CPU is a future PyTorch feature coming in 1.10

[edgarriba]

Closing this – unless you have a kornia specific non-working code please reopen. Otherwise touch base the lightning team in their channels.