Source code for kornia.augmentation._2d.geometric.elastic_transform

from typing import Any, Dict, Optional, Tuple, Union

import torch
from torch import Tensor

from kornia.augmentation._2d.geometric.base import GeometricAugmentationBase2D
from kornia.constants import Resample
from kornia.geometry.transform import elastic_transform2d


[docs]class RandomElasticTransform(GeometricAugmentationBase2D): r"""Add random elastic transformation to a tensor image. .. image:: _static/img/RandomElasticTransform.png Args: kernel_size: the size of the Gaussian kernel. sigma: The standard deviation of the Gaussian in the y and x directions, respectively. Larger sigma results in smaller pixel displacements. alpha: The scaling factor that controls the intensity of the deformation in the y and x directions, respectively. align_corners: Interpolation flag used by `grid_sample`. resample: Interpolation mode used by `grid_sample`. Either 'nearest' (0) or 'bilinear' (1). mode: Deprecated: Interpolation mode used by `grid_sample`. Either 'bilinear' or 'nearest'. Please use the `resample` argument instead. padding_mode: The padding used by ```grid_sample```. Either 'zeros', 'border' or 'refection'. same_on_batch: apply the same transformation across the batch. p: probability of applying the transformation. keepdim: whether to keep the output shape the same as input (True) or broadcast it to the batch form (False). .. note:: This function internally uses :func:`kornia.geometry.transform.elastic_transform2d`. Examples: >>> import torch >>> img = torch.ones(1, 1, 2, 2) >>> out = RandomElasticTransform()(img) >>> out.shape torch.Size([1, 1, 2, 2]) To apply the exact augmenation again, you may take the advantage of the previous parameter state: >>> input = torch.randn(1, 3, 32, 32) >>> aug = RandomElasticTransform(p=1.) >>> (aug(input) == aug(input, params=aug._params)).all() tensor(True) """ def __init__( self, kernel_size: Tuple[int, int] = (63, 63), sigma: Tuple[float, float] = (32.0, 32.0), alpha: Tuple[float, float] = (1.0, 1.0), align_corners: bool = False, resample: Union[str, int, Resample] = Resample.BILINEAR.name, padding_mode: str = "zeros", same_on_batch: bool = False, p: float = 0.5, keepdim: bool = False, return_transform: Optional[bool] = None, ) -> None: super().__init__( p=p, return_transform=return_transform, same_on_batch=same_on_batch, p_batch=1.0, keepdim=keepdim ) self.flags = dict( kernel_size=kernel_size, sigma=sigma, alpha=alpha, align_corners=align_corners, resample=Resample.get(resample), padding_mode=padding_mode, ) def generate_parameters(self, shape: torch.Size) -> Dict[str, Tensor]: B, _, H, W = shape if self.same_on_batch: noise = torch.rand(1, 2, H, W, device=self.device, dtype=self.dtype).expand(B, 2, H, W) else: noise = torch.rand(B, 2, H, W, device=self.device, dtype=self.dtype) return dict(noise=noise * 2 - 1) # TODO: It is incorrect to return identity def compute_transformation(self, input: Tensor, params: Dict[str, Tensor], flags: Dict[str, Any]) -> Tensor: return self.identity_matrix(input) def apply_transform( self, input: Tensor, params: Dict[str, Tensor], flags: Dict[str, Any], transform: Optional[Tensor] = None ) -> Tensor: return elastic_transform2d( input, params["noise"].to(input), flags["kernel_size"], flags["sigma"], flags["alpha"], flags["align_corners"], flags["resample"].name.lower(), flags["padding_mode"], )