Optional spline interpolation on shapes

🚀 Feature

Shapes could have the option of a smooth interpolation between points.

Motivation

In my image analysis pipeline, I make use of smooth curves which interpolate through points. This functionality is also available in e.g. FIJI/ImageJ.

Pitch

My initial idea is that in addition to the opacity/edge width sliders and the blending/face/edge color dropdowns, there could be a “smooth” checkbox which would enable cubic interpolation when checked, and linear interpolation when unchecked. Alternatively, there could be a dropdown menu with different interpolation options.

Alternatives

These smooth versions could be their own shapes? That seems less elegant though.

Additional context

This interpolation is actually quite easy thanks to SciPy. I made a rough proof-of-concept by calling _set_meshes with the interpolated data within _update_displayed_data in napari/layers/shapes/_shapes_models/path.py:

def _update_displayed_data(self):
        """Update the data that is to be displayed."""
        # For path connect every all data

        if self.data_displayed.shape[0] > 2:
            distance = np.cumsum(np.sqrt(np.sum(np.diff(self.data_displayed[:-1,:], axis=0)**2, axis=1 )) )
            distance = np.insert(distance, 0, 0)/distance[-1]

            # the number of sampled data points might need to be carefully thought
            # about (might need to change with image scale?)
            alpha = np.linspace(0, 1, 75)

            spl = CubicSpline(distance, self.data_displayed[:-1, :])
            interp_pts = spl(alpha)

            self._set_meshes(interp_pts, face=False, closed=False)
        else:
            self._set_meshes(self.data_displayed, face=False, closed=False)

        self._box = create_box(self.data_displayed)

        data_not_displayed = self.data[:, self.dims_not_displayed]
        self.slice_key = np.round(
            [
                np.min(data_not_displayed, axis=0),
                np.max(data_not_displayed, axis=0),
            ]
        ).astype('int')

I recorded a short video of this in action

I’m not really familiar with napari’s architecture, so this is all I’ve done so far, but if the above proposal sounds interesting, I’d be happy to keep working on this.

[half-adder]

Excited that this has some interest!

@tlambert03 I do have code that can measure along these curves - I use them to make measurements in my microscopy. Here’s the function I’m using to do it.

Basically i just use np.map_coordinates to sample the image along the points of the spline.

If you want the avg. intensity around the shape, it does get a bit more complex (although there might be a much easier way than what I’m doing!). I think if the shape has been given width through the UI it makes sense to do.

For this, I just calculate the normal lines at each sampled point along the curve, then use np.map_coordinates to measure under those, resulting in a measurement rectangle (which is essentially a “straightened” image along the curve). Then you can flatten it with the average. In that sample code, I actually weight the average around a gaussian centered on the curve, but that is specific to my needs, and I’m guessing is needlessly complex for this package.

[sofroniewn]

See here for an explanation of what might be required to finish the work https://github.com/napari/napari/pull/2345#issuecomment-955157371