Although deep learning models are increasingly central to image analysis, their application remains uneven and difficult, particularly when using the latest models for large data. There are a lot of barriers to use, mainly around complex installation and difficulty in configuration, for which each new model can bring new challenges. This is made more complicated when trying to use such models on HPC. AI OnDemand (AIoD) was designed to make these models more accessible, scalable, and reproducible. By separating out the user interface from the computation, researchers can access models through familiar interfaces such as Napari, but run deep learning models on any compute while the installation and execution is handled by our Nextflow pipeline (which parallelises running these models as much as possible!). AIoD is designed to be easily extendable and user-friendly, making it as useful to non-computational microscopists as it is to AI model developers!

In this talk I will: give an overview of AIoD; illustrate the different ways how to use it; how it can be used at different institutes; and, how different members of the community can benefit from it and even contribute to it!