Transformations

Modules

Add Noise

The Add_Noise is often used for data augmentation and sometimes as regularization.
The function adds gaussian centered around the given mean and standard deviation to the input.

Parameters: inputs (Torch.Tensor,required): The input tensor which has to be transformed. mean (Torch.Tensor,default = None): The mean of gaussian noise to be added as a tensor. std (Torch.Tensor,default = None): The std of gaussian noise to be added as a tensor.

Usage:

import torch
input = torch.ones(64,512,32,32,32)
from niftytorch.Transformations.Transformations import Add_Noise
output = Add_Noise(input,mean = torch.zeros(input.shape),std = torch.eye(input.shape))

Rotate 90

Rotate the input tensor by 90.

Parameters: inputs (torch.Tensor,required): The input tensor which needs to rotated.

Usage:

import torch
input = torch.ones(64,512,32,32,32)
from niftytorch.Transformations.Transformations import Rotate_90
output = Rotate_90(input)

Rotate 180

Rotate the input tensor by 180.

Parameters: inputs (torch.Tensor,required): The input tensor which needs to rotated.

Usage:

import torch
input = torch.ones(64,512,32,32,32)
from niftytorch.Transformations.Transformations import Rotate_180
output = Rotate_180(input)

Rotate 270

Rotate the input tensor by 270.

Parameters: inputs (torch.Tensor,required): The input tensor which needs to rotated.

Usage:

import torch
input = torch.ones(64,512,32,32,32)
from niftytorch.Transformations.Transformations import Rotate_270
output = Rotate_270(input)

Random_Segmentation_Crop

The Random_Segmentation_Crop is often used with segmentation.
The idea behind random crop is select regions around lesions and use only this region to train the network.

Parameters: input: the input 3D image. mask: the mask of the lesion region. context: the region around the lesion to be considered for crop.

Usage:

import torch
input = torch.ones(64,512,32,32,32)
mask = torch.zeros(64,512,32,32,32)
context = 32
from niftytorch.Transformations.Transformations import Random_Segmentation_Crop
input,mask = Random_Segmentation_Crop(input,mask,context)

Resize

The Resize module is used to resize the input tensor to the given size.

Parameters: input (torch.Tensor,required): the input 3D image. mask (torch.Tensor,default = None): the mask of the lesion region. common: the size to which the tensor is to be resized.

Usage:

import torch
input = torch.ones(64,512,32,32,32)
mask = torch.zeros(64,512,32,32,32)
context = 32
from niftytorch.Transformations.Transformations import Resize
input,mask = Resize(input,mask,context)