Source code for kornia.augmentation._3d.geometric.crop

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

from torch import Tensor
from torch.nn.functional import pad

from kornia.augmentation import random_generator as rg
from kornia.augmentation._3d.base import AugmentationBase3D
from kornia.constants import Resample
from kornia.geometry import crop_by_transform_mat3d, get_perspective_transform3d

[docs]class RandomCrop3D(AugmentationBase3D): r"""Apply random crop on 3D volumes (5D tensor). Crops random sub-volumes on a given size. Args: p: probability of applying the transformation for the whole batch. size: Desired output size (out_d, out_h, out_w) of the crop. Must be Tuple[int, int, int], then out_d = size[0], out_h = size[1], out_w = size[2]. padding: Optional padding on each border of the image. Default is None, i.e no padding. If a sequence of length 6 is provided, it is used to pad left, top, right, bottom, front, back borders respectively. If a sequence of length 3 is provided, it is used to pad left/right, top/bottom, front/back borders, respectively. pad_if_needed: It will pad the image if smaller than the desired size to avoid raising an exception. Since cropping is done after padding, the padding seems to be done at a random offset. fill: Pixel fill value for constant fill. Default is 0. If a tuple of length 3, it is used to fill R, G, B channels respectively. This value is only used when the padding_mode is constant. padding_mode: Type of padding. Should be: constant, edge, reflect or symmetric. Default is constant. resample: resample mode from "nearest" (0) or "bilinear" (1). return_transform: if ``True`` return the matrix describing the transformation applied to each input tensor. If ``False`` and the input is a tuple the applied transformation won't be concatenated same_on_batch: apply the same transformation across the batch. align_corners: interpolation flag. keepdim: whether to keep the output shape the same as input (True) or broadcast it to the batch form (False). Shape: - Input: :math:`(C, D, H, W)` or :math:`(B, C, D, H, W)`, Optional: :math:`(B, 4, 4)` - Output: :math:`(B, C, , out_d, out_h, out_w)` Note: Input tensor must be float and normalized into [0, 1] for the best differentiability support. Additionally, this function accepts another transformation tensor (:math:`(B, 4, 4)`), then the applied transformation will be merged int to the input transformation tensor and returned. Examples: >>> import torch >>> rng = torch.manual_seed(0) >>> inputs = torch.randn(1, 1, 3, 3, 3) >>> aug = RandomCrop3D((2, 2, 2), p=1.) >>> aug(inputs) tensor([[[[[-1.1258, -1.1524], [-0.4339, 0.8487]], <BLANKLINE> [[-1.2633, 0.3500], [ 0.1665, 0.8744]]]]]) To apply the exact augmenation again, you may take the advantage of the previous parameter state: >>> input = torch.rand(1, 3, 32, 32, 32) >>> aug = RandomCrop3D((24, 24, 24), p=1.) >>> (aug(input) == aug(input, params=aug._params)).all() tensor(True) """ def __init__( self, size: Tuple[int, int, int], padding: Optional[Union[int, Tuple[int, int, int], Tuple[int, int, int, int, int, int]]] = None, pad_if_needed: Optional[bool] = False, fill: int = 0, padding_mode: str = "constant", resample: Union[str, int, Resample] =, return_transform: Optional[bool] = None, same_on_batch: bool = False, align_corners: bool = True, p: float = 1.0, keepdim: bool = False, ) -> None: # Since PyTorch does not support ragged tensor. So cropping function happens batch-wisely. super().__init__( p=1.0, return_transform=return_transform, same_on_batch=same_on_batch, p_batch=p, keepdim=keepdim ) self.flags = dict( size=size, padding=padding, pad_if_needed=pad_if_needed, padding_mode=padding_mode, fill=fill, resample=Resample.get(resample), align_corners=align_corners, ) self._param_generator = cast(rg.CropGenerator3D, rg.CropGenerator3D(size, None)) def precrop_padding(self, input: Tensor, flags: Optional[Dict[str, Any]] = None) -> Tensor: flags = self.flags if flags is None else flags padding = flags["padding"] if padding is not None: if isinstance(padding, int): padding = [padding, padding, padding, padding, padding, padding] elif isinstance(padding, (tuple, list)) and len(padding) == 3: padding = [padding[0], padding[0], padding[1], padding[1], padding[2], padding[2]] elif isinstance(padding, (tuple, list)) and len(padding) == 6: padding = [padding[0], padding[1], padding[2], padding[3], padding[4], padding[5]] # type: ignore else: raise ValueError(f"`padding` must be an integer, 3-element-list or 6-element-list. Got {padding}.") input = pad(input, padding, value=flags["fill"], mode=flags["padding_mode"]) if flags["pad_if_needed"] and input.shape[-3] < flags["size"][0]: padding = [0, 0, 0, 0, flags["size"][0] - input.shape[-3], flags["size"][0] - input.shape[-3]] input = pad(input, padding, value=flags["fill"], mode=flags["padding_mode"]) if flags["pad_if_needed"] and input.shape[-2] < flags["size"][1]: padding = [0, 0, (flags["size"][1] - input.shape[-2]), flags["size"][1] - input.shape[-2], 0, 0] input = pad(input, padding, value=flags["fill"], mode=flags["padding_mode"]) if flags["pad_if_needed"] and input.shape[-1] < flags["size"][2]: padding = [flags["size"][2] - input.shape[-1], flags["size"][2] - input.shape[-1], 0, 0, 0, 0] input = pad(input, padding, value=flags["fill"], mode=flags["padding_mode"]) return input def compute_transformation(self, input: Tensor, params: Dict[str, Tensor], flags: Dict[str, Any]) -> Tensor: transform: Tensor = get_perspective_transform3d(params["src"].to(input), params["dst"].to(input)) transform = transform.expand(input.shape[0], -1, -1) return transform def apply_transform( self, input: Tensor, params: Dict[str, Tensor], flags: Dict[str, Any], transform: Optional[Tensor] = None ) -> Tensor: transform = cast(Tensor, transform) return crop_by_transform_mat3d( input, transform, flags["size"], mode=flags["resample"].name.lower(), align_corners=flags["align_corners"] ) def forward(self, input: Tensor, params: Optional[Dict[str, Tensor]] = None, **kwargs) -> Tensor: # type: ignore # TODO: need to align 2D implementations input = self.precrop_padding(input) return super().forward(input, params) # type:ignore