Aydin Use-Cases, and the Art & Science of Image Denoising.
While designing Aydin, we came to the realisation that there is no silver bullet, there is not a single algorithm nor single set of parameters that can reliably denoise all images. More fundamentally, ‘image denoising’ is not a singular problem, but a multiplicity of challenges. There are many different kinds of image noises and related degradations and it requires a lot of know-how to both understand this diversity and know what to do for each kind of noise. This diversity comes from the inherent diversity of imaging modalities: different instruments, optics, detectors, contrast mechanisms, photo-chemistry, etc… This extends well beyond just optical images into other imaging modalities – it essentially holds true for any mechanism that generates images and more generally any multi-dimensional array-like measurement.
The use-cases presented below are the ideal starting point to understand how to choose among denoising algorithms, among pre- and post- processing steps, and how to adjust their parameters. We are actively populating this list with more datasets, and further improving this material, please check this page for updates!
Spinning-Disk Confocal Images of Zebrafish Embryos from Royer Lab (CZ Biohub, San Francisco)
Spinning-Disk Confocal Microscopy Images of Mouse Embryos from the Maitre Lab (Curie, Paris)
Chicken Embryos LSM 780 Images from the Pourquie lab (Harvard, Boston)
Note: We are always interested in learning about new challenging images, please contact us by filling an issue here if you face difficulties or have a dataset that resists our methods.