Can't run Figure.export in an offscreen setting.

[jgleyze]

I am running fastplotlib (v0.5.1) in a headless server and I'd like to export the render of my figure however I can't seem to make the Figure.export method work...

Here is a small reproduction of my problem:

import fastplotlib as fpl
import numpy as np

if __name__ == "__main__":
    fig = fpl.Figure(canvas="offscreen", size=(900, 600))
    fig[0, 0].add_image(np.random.rand(100, 100, 3))
    rgb = fig.export_numpy(rgb=True)

This throws AttributeError: 'Ordered2FragmentBlender' object has no attribute 'color_tex'

I might forget something, I am quite new to fpl !

[kushalkolar]

Hi thanks for posting. You need to manually perform the render calls when using the offscreen canvas, here's an example from our tests:

fastplotlib/examples/tests/test_examples.py

Lines 103 to 125 in 0293fa3

	if fpl.IMGUI:
	# there doesn't seem to be a resize event for the manual offscreen canvas
	example.figure.imgui_renderer._backend.io.display_size = (
	example.figure.canvas.get_logical_size()
	)
	# run this once so any edge widgets set their sizes and therefore the subplots get the correct rect
	# hacky but it works for now
	example.figure.imgui_renderer.render()
	example.figure._fpl_reset_layout()
	# render each subplot
	for subplot in example.figure:
	subplot.viewport.render(subplot.scene, subplot.camera)
	# flush pygfx renderer
	example.figure.renderer.flush()
	if fpl.IMGUI:
	# render imgui
	example.figure.imgui_renderer.render()
	# render a frame
	img = np.asarray(example.figure.renderer.target.draw())

By the way what is your use case here? The remote frame buffer is usually what you want if you're doing offscreen rendering.

[jgleyze]

Cool, thank you ! You are right I am in fact using the export function rather than the export_numpy, since it was the latter that crashed I put it in my issue. My use case is quite niche, I am using fpl for very fast plotting data from of a simulated robotic environment and then send the plot as an image to a web app so that I can monitor real time at high frequency what's happening remotely.

[kushalkolar]

jupyter_rfb would probably suite your use case better