GeezerVison

took some 3d movie gen code from caiman to generate a 3d movie:

@FlynnOConnell

File [~\repos\fastplotlib\fastplotlib\graphics\features\_image.py:147](http://localhost:8888/lab/tree/demos/notebooks/~/repos/fastplotlib/fastplotlib/graphics/features/_image.py#line=146), in TextureArray._fix_data(self, data)
    145 def _fix_data(self, data):
    146     if data.ndim not in (2, 3):
--> 147         raise ValueError(
    148             "image data must be 2D with or without an RGB(A) dimension, i.e. "
    149             "it must be of shape [rows, cols], [rows, cols, 3] or [rows, cols, 4]"
    150         )
    152     # let's just cast to float32 always
    153     return data.astype(np.float32)

ValueError: image data must be 2D with or without an RGB(A) dimension, i.e. it must be of shape [rows, cols], [rows, cols, 3] or [rows, cols, 4]

data = np.random.randn(50, 2, 448, 448)
data.shape
>> (50, 2, 448, 448)

The ability to view a time-series with a 4th dimension would be nice here

Also how difficult is something like scale=[1, 4, 1] to account for anisotropy in XY and Z? Here I have a volume of [14, 448, 448] ZXY, but from the side its a flat plane. This dataset is [2um x 2um x 17um]

Also how difficult is something like scale=[1, 4, 1] to account for anisotropy in XY and Z? Here I have a volume of [14, 448, 448] ZXY, but from the side its a flat plane. This dataset is [2um x 2um x 17um]

nevermind :)

volume.world_object.world.scale_z = 15.0

Some other helpful tidbits:

hlut = fpl.HistogramLUTTool(volume.data.value, volume)
volume.interpolation = "linear"
volume.cmap = "gnuplot2"

fig[0, 0].docks["right"].add_graphic(hlut)
fig[0, 0].docks["right"].size = 80

fig[0, 0].docks["right"].controller.enabled = False
fig[0, 0].docks["right"].camera.maintain_aspect = False
fig[0, 0].docks["right"].auto_scale(maintain_aspect=False)

@almarklein are RGB volumes supported? Is it even a thing?

Yes, looks like it is; it uses the same logic (in the shader) as the image. I have never encountered it in the wild though.

ok all the basics are down, switchable materials, can change material props:

Need to figure out why regex is turning VolumeSliceMaterial into slic instead of slice

In other news, the latest pycharm is full of bugs, wtf is this. Also it wanted to auto insert the old rst structured docstring template even though I configured it to do numpydoc format 🤮

All that's left is tests, 2D over time example, and RGB(A) examples. I also think I want to merge #849 and then rebase this so the examples will be much simpler.

Also pygfx has non-orthogonal slicing:

ok there's a bug in the TextureArray or tiling somewhere, because when a dim is greater than the limit it is tiling along the wrong dim 😆

The rendering at the edges is weird if one of the dims is larger than the 3d texture size limit so let's just not support this, I think the uses are very much rare edge cases. Also if the 3d texture size limit is 16,383, then 16384x16384x16384 with float32 is 17 TB which is more than any GPU that I know of can handle 😂 .

📚 Docs preview built and uploaded! https://www.fastplotlib.org/ver/volume-image

More tests for the TextureArray for volumes would be nice, but can do that later. I think tiling of volume is a rare edge case so if there is a bug it will affect very rare usecases not already covered by the tests.

Amazing! @kushalkolar I'll test this out!

@clewis7 gtg! 🥳 and then we can merge #904 once I make new ground truth screenshots for that and then #883