RGB volumes

[VolkerH]

Hi,

excuse me if this is a silly question but I've been going through the example notebooks and the documentation and I haven't been able to figure out how to render RGB volumes, e.g. a numpy array of shape [nx, ny, nz, 3] where the last dimensions are RGB values.

If I interpret the examples correctly, most examples seem to plot single-valued volume data and apply some sort of intensity-based RGB lookup table.

[maartenbreddels]

Indeed that is usually the case, still, interesting question.
However, do you mean [nx, ny, nz, 3] or [nx, ny, nz, 4], since without an opacity it would always render an opaque cube.

[VolkerH]

Actually the data I have is microscopy fluorescence intensity data in multiple channels. I usually want to visualize up to three channels of them simultaneously, typically assigning one of the R,G,B channels to each of them.
The arrays I work with therefore don't have an explicit opacity channel. That would have to be computed, maybe as the inverse of the total intensity in all channels. The (voxels that don't have fluorescence emission are transparent).

[maartenbreddels]

I'm working on multivolume rendering now, which can be completely overlapping, half overlapping or non overlapping. Is it an option for you to have 3 volumes, each with a single channel, and each with its own transfer function?
If the values really represent RGB values directly, you can have each of the transfer functions do exactly that.

[GenevieveBuckley]

Hi @maartenbreddels, I have the same use case as Volker so I'm going to jump in here.

The kind of microscopy image data we'd be working with would generally arrive as numpy arrays with shape (plane, row, column, channel). Plane, row, and column are the three spatial dimensions (generally called z, y, and x). Channel is represented by color in the image volume. Note that you can have lots of channels - we're not just restricted to 3 for red, green and blue. Alpha is never specified in the data itself, that's something that's part of the particular transfer function you'd apply.

The multivolume rendering #143 should make it totally feasible to display this kind of data. We could slice on the channel axis, and plot separate overlapping volumes with a different transfer function for each.

I haven't been able to get the branch with multivolume rendering working yet but if you can help me fix that, I can work on a pull request that does what I've described here.

[maartenbreddels]

I think this can now be done using multivolume rendering. Render each channel as a separate volume, and give them 'ramp' transfer function, having a constant color (say red for the first, green for the second), and opacity increasing from 0 to 1.

[GenevieveBuckley]

With a bit of fiddling I made this:

I'm planning to clean the code up a bit and submit a pull request soon-ish.

[maartenbreddels]

Really nice! Can you open a PR so I can take a look at how you did it? (don't worry about the code not being polished yet).