Applications#

Click on your algorithm of choice to explore various applications. We collected the algorithms based on the type of training data they require!

Keywords#

• no ground-truth: The algorithm trains without clean images.
• single image: The algorithm can train on a single image.
• pairs of noisy images: The algorithm requires pairs of noisy images.
• ground-truth: The algorithm requires pairs of clean and noisy images.

Denoising noisy images without clean data#

You have noisy images and no clean images? No problem! These algorithms can help you, as they do not require any ground-truth data. You can also train on a single image of reasonable size.

If you have multiple noisy instances of the same structure (e.g. a noisy time-lapse), then Noise2Noise might be the right choice for you.

Supervised restoration with clean images#

If you have pairs of clean (e.g. high SNR, long exposure or high laser power) and noisy images, then CARE might be the right choice for you.

Note that CARE can be used for a variety of tasks, such as denoising, deconvolution, isotropic resolution restoration or projection.

Using the Lightning API#

If you need more control on the algorithm training, for instance to implement or replace features, you can use the Lightning API.

It uses PyTorch Lightning and the CAREamics Lightning components.