Support modes other than nearest neighbour downsampling for HeatmapVis

[stufisher]

For diffraction applications it would be nice to support some sort of filtering beyond nearest neighbour. If you have a diffraction image being displayed at say 1/4 of its size with single pixel diffraction spots you basically cant see anything in <HeatmapVis>. I know @t20100 recently implemented this in silx, i don't know how much work it is for h5web.

[loichuder]

i don't know how much work it is for h5web.

Not much, I expect. We should only need to expose minFilter (https://threejs.org/docs/index.html#api/en/textures/DataTexture) up to the needed component.

@stufisher Are you using the HeatmapVis component ? Or a lower/higher component ?

[stufisher]

I'm using both HeatmapVis and h5web/app

[loichuder]

So we would need to expose it all the way to HeatmapVis. Unless you have another idea @axelboc ?

[axelboc]

• Adding a prop of sorts to HeatmapVis should be doable.
• For @h5web/app, we're back to the same architectural challenge of allowing the visualizations to be customized/configured. Or we'd have to add a control to the Heatmap toolbar, which has UX implications obviously.

[stufisher]

<HeatmapVis> alone would already be really useful i think

[t20100]

Using a linear minFilter should improve the display but peaks will also disappear at some point.

What is done in silx is a bit different: it takes the maximum (or mean/minimum) of the data points that are inside each pixel of the screen as the displayed value. It's a sort of 2D histogram where each pixel of the screen is a bin.
It's done in Python (so in the case of silx it also works with matplotlib:
https://github.com/silx-kit/silx/blob/a9eddc5917d0d0b5c774d7cc4df3a074cf538ee6/src/silx/gui/plot/items/image_aggregated.py#L105-L129
but that should be possible to do a shader for this as well.