class DeepImagePrior[source]
DeepImagePrior(shape:list,n_channels:int=1,n_inital_channels:int=4) ::Module
The deep image prior (DIP) [1] is a type of convolutional neural network used to enhance a given image with no prior training data other than the image itself. A neural network is randomly initialized and used as prior to solve inverse problems such as noise reduction, super-resolution, and inpainting. Image statistics are captured by the structure of a convolutional image generator rather than by any previously learned capabilities.
| Type | Default | Details | |
|---|---|---|---|
shape |
list |
A list containing three entries that define the number of voxels in each direction. | |
n_channels |
int |
1 |
The number of input channels. |
n_inital_channels |
int |
4 |
T he number of channels after the first encoding block. The model has a total 4 encoding and 4 decoding blocks, and the number of channels is doubled in each encoding step. |
DeepImagePrior.forward[source]
DeepImagePrior.forward()
The forward pass of the DIP with a fixed random noise input. Returns a torch.Tensor object.
[1] Ulyanov, Dmitry, Andrea Vedaldi, and Victor Lempitsky. "Deep image prior." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.