Utils#

Plotting#

We provide some plotting functions that are adapted to inverse problems. The main plotting function is deepinv.utils.plot, which can be used to quickly plot a list of tensor images.

>>> from deepinv.utils import plot
>>> import torch
>>> x1 = torch.rand(4, 3, 16, 16)
>>> x2 = torch.rand(4, 3, 16, 16)
>>> plot([x1, x2], titles=['x1', 'x2'])

We provide other plotting functions that are useful for inverse problems:

Table 29 Utility Functions and Descriptions#
Function	Description
`deepinv.utils.plot()`	Plots a list of tensor images with optional titles.
`deepinv.utils.plot_curves()`	Plots curves for visualizing metrics over optimization iterations.
`deepinv.utils.plot_parameters()`	Visualizes model parameters over optimization iterations.
`deepinv.utils.plot_inset()`	Plots a list of images with zoomed-in insets extracted from the images.
`deepinv.utils.plot_videos()`	Plots and animates a list of image sequences.
`deepinv.utils.plot_ortho3D()`	Plots 3D orthographic projections for analyzing data or model outputs in three dimensions.

Logging#

Table 30 Logging functionality#
Function/class	Description
`deepinv.utils.AverageMeter()`	Store values and keep track of average and std.
`deepinv.utils.get_timestamp()`	Get current timestamp string.

Other#

We provide some useful utility and demo functions:

Table 31 Utility and demo Functions#
Function	Description
`deepinv.utils.get_freer_gpu()`	Finds the GPU with the most available memory.
`deepinv.utils.get_data_home()`	Get the path to the default directory for storing datasets.
`deepinv.utils.get_image_url()`	Get URL for image from DeepInverse HuggingFace repository.
`deepinv.utils.get_degradation_url()`	Get URL for degradation from DeepInverse HuggingFace repository.
`deepinv.utils.load_url_image()`	Loads an image directly from a URL for experiments or demos.
`deepinv.utils.load_image()`	Loads a local image file for experiments or demos.
`deepinv.utils.load_dataset()`	Loads an ImageFolder dataset from DeepInverse HuggingFace repository.
`deepinv.utils.load_degradation()`	Loads a degradation tensor from DeepInverse HuggingFace repository.

TensorList#

The deepinv.utils.TensorList class is a wrapper around a list of tensors. It allows performing elementary operations on the list of tensors, such as sum, multiplication, etc.:

>>> from deepinv.utils import TensorList
>>> import torch
>>> x1 = torch.ones(2, 3, 2, 2)
>>> x2 = torch.ones(2, 1, 3, 3)
>>> t1 = TensorList([x1, x2])
>>> t2 = TensorList([x1*2, x2/2])
>>> t3 = t1 + t2