napari: an n-dimensional image viewer in Python (Vis2021)

I recorded this talk for an audience consisting mainly of biologists and bioimage analysts, rather than Python experts. It gives a fast introduction to the state of napari in early 2021. Full abstract below.



napari (https://napari.org) is a new(ish) n-dimensional array viewer for Python. It was designed to be n-dimensional first, blazing fast, and play well with the existing scientific Python ecosystem.


napari’s main architectural choices are:
– a 2D or 3D canvas that is powered by VisPy and OpenGL for buttery-smooth interaction with the current view,
– slicing sliders for the axes not currently being displayed: as many are generated as needed for the current dataset, allowing exploration of higher-dimensional data, such as multi-channel, volumetric 3D time series,
– a *layer list*, to enable overlays of different views of the same data space (such as segmentations of images or point detections)
– simple data models, enabling two-way data flow between display, annotations, and analysis results.


I will demonstrate how these principles enable diverse workflows, such as:
– viewing n-dimensional images (NumPy arrays) and larger-than-RAM datasets using dask and zarr
– adding interactive annotation steps within custom computational workflows
– creating a quick viewer for custom images using plugins
– viewing neural network results during training on the fly


napari was developed as an open source project from its inception, and over 70 people have contributed to its development thus far.

• #people_blogs
• #napari
• #python
• #scipy
• #numpy
• #dask