Applying successives filters / function on the same layer

❓ Questions and Help

First of all, thank you for magicgui, just started playing with it and it’s pretty awesome. I have a pretty simple question:

If I take the example from the documentation:

@magicgui(
        auto_call=True,
        sigma={"widget_type": QDoubleSlider, "maximum": 6, "fixedWidth": 400},
        mode={"choices": ["reflect", "constant", "nearest", "mirror", "wrap"]},
    )
    def gaussian_blur(layer: Image, sigma: float = 1.0, mode="nearest") -> Image:
        """Apply a gaussian blur to ``layer``."""
        if layer:
            return skimage.filters.gaussian(layer.data, sigma=sigma, mode=mode)

But let say I wanted to apply skimage.filters.median as well to the same layer image while being able to adjust the selem size, what would be the best and cleanest way to go about it? I try few things but ended work on 2 separates layers.

[ ] I have searched the documentation

[tlambert03]

The core idea here is that magicgui basically just brokers mappings between types and widgets (in the function signature), or return annotations and callbacks. magicgui doesn’t know anything about napari, (it doesn’t import napari or depend on napari in any way), but it does allow third party packages to “register” their own custom classes with magicgui so as to provide “sane defaults”: for instance, when you annotate a function parameter as a napari Layer def function(image1: napari.layers.Image):, then magic gui will render that as a dropdown box with all of the current viewer image layers because napari has told it to do so. Had napari not done that, then magicgui would have no idea what to do with a parameter of type napari.layers.Image (unless the user instructed it manually).

If at any point you want to “opt-out” of the behavior that a third party package like napari has registered, then you either can not annotate your parameter or return type as that third party class … in which case you’ll need to tell magicgui what to do with that type, by either directly specifying the widget_type argument (manually passing a QtWidget for magicgui to use), or you can override the third party behavior by registering your own widget for the corresponding type:

from magicgui import register_type

register_type(napari.layers.Layer, widget_type=MyCustomQtWidget)

[tlambert03]

That’s definitely a use case I’d like to support here! Ultimately, I’d like these widgets to be composable, and nestable (e.g. using a function as a parameter makes a nested widget).

For what it’s worth, I had a project that I was working on for a while that attempts to build that sort of “autogui” from an arbitrary sequence of python functions: mosaicpy, (see the “basic idea” in the readme there). I think it would be possible here to do something similar: make a “list item widget” where the items in the list are (sortable) functions, each of which is rendered as a magicgui, and outputs are chained together…

Thanks for your note! I’ll definitely keep this use case in mind and try to work on something soon.