Camera3D Improvements

[ffreyer]

Description

Some general camera improvements/changes

Breaking

• revert the switch to fov based zoom. Camera3D now uses the same zoom style as old_cam3d! again, i.e. translating eyeposition.
  • to keep clipping issues (Zooming without entering shape #602, Camera weirdness #941) from occurring near and far are now relative to norm(eyeposition - lookat), i.e. the zoom level.
  • because of this stretch/contract view hotkeys have been removed as they are equivalent to zoom in/out. The old zoom in/out is now available as increase/decrease fov.
• Separate attributes into cam.controls for hotkeys and cam.settings for everything else

Non-breaking

• make near & far relative to norm(eyeposition - lookat), i.e. zoom level
• scale translations with zoom level and fov. This normalizes the perceived translation speed
• scale camera centering on bounding boxes with fov. This makes the full bbox visible at a larger range of fovs.
• add orbital `update_cam!(scene, cam, phi, theta[, radius, center])
• add deprecation warning for old_cam3d! etc. Camera3D can now reproduce this, though not exactly due to QoL tweaks.
• improve default hotkey choices
• add functionality to focus camera on a point on a plot

Fixes #2721, #1600, #1236 (y/z swap, documentation)

DataInspector fixes

I moved a bunch of code around that's also used by DataInspector so I did some testing and fixing on that too. Most of these fixes are probably not related to this pr.

• fixed error related to getindex_sv(::Range, idx)
• fixed blinking in image inspection
• fixed typos in arrows (closes callback error in arrows data inspection #2764) & lines(egments)
• fixed incorrect values when inspecting volumeslices
• fixed incorrect indicator placement in lines and added workaround to avoid multiple indicators

Type of change

Breaking change

Checklist

• Added an entry in NEWS.md (for new features and breaking changes)
• Added or changed relevant sections in the documentation
• Check DataInspector
• Document changes to ispressed

[MakieBot]

Compile Times benchmark

Note, that these numbers may fluctuate on the CI servers, so take them with a grain of salt. All benchmark results are based on the mean time and negative percent mean faster than the base branch. Note, that GLMakie + WGLMakie run on an emulated GPU, so the runtime benchmark is much slower. Results are from running:

using_time = @ctime using Backend
# Compile time
create_time = @ctime fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @ctime Makie.colorbuffer(display(fig))
# Runtime
create_time = @benchmark fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @benchmark Makie.colorbuffer(display(fig))

	using	create	display	create	display
GLMakie	13.80s (13.40, 14.94) 0.52+-	1.25s (1.19, 1.36) 0.05+-	898.28ms (851.63, 968.95) 45.53+-	15.99ms (15.63, 16.51) 0.31+-	186.82ms (184.07, 191.12) 2.71+-
master	13.66s (13.46, 14.02) 0.18+-	1.26s (1.24, 1.34) 0.04+-	888.78ms (861.61, 910.68) 14.67+-	15.92ms (15.70, 16.08) 0.13+-	185.27ms (183.09, 191.02) 2.81+-
evaluation	+1.01%, 0.14s invariant (0.36d, 0.52p, 0.35std)	-0.95%, -0.01s invariant (-0.26d, 0.64p, 0.04std)	+1.06%, 9.51ms invariant (0.28d, 0.61p, 30.10std)	+0.43%, 0.07ms invariant (0.29d, 0.60p, 0.22std)	+0.83%, 1.54ms invariant (0.56d, 0.32p, 2.76std)
CairoMakie	9.73s (9.58, 9.80) 0.09+-	1.03s (1.00, 1.04) 0.01+-	222.56ms (219.16, 225.33) 2.13+-	9.59ms (9.47, 9.70) 0.09+-	5.72ms (5.59, 5.80) 0.08+-
master	9.72s (9.47, 9.88) 0.15+-	1.02s (0.99, 1.04) 0.02+-	234.61ms (229.58, 238.54) 3.35+-	9.61ms (9.49, 9.76) 0.11+-	5.77ms (5.66, 5.91) 0.09+-
evaluation	+0.09%, 0.01s invariant (0.07d, 0.89p, 0.12std)	+0.40%, 0.0s invariant (0.26d, 0.63p, 0.02std)	-5.41%, -12.04ms faster✅ (-4.29d, 0.00p, 2.74std)	-0.27%, -0.03ms invariant (-0.25d, 0.64p, 0.10std)	-0.74%, -0.04ms invariant (-0.48d, 0.39p, 0.09std)
WGLMakie	15.54s (15.23, 15.91) 0.21+-	1.42s (1.36, 1.48) 0.04+-	14.09s (13.92, 14.24) 0.11+-	16.46ms (16.10, 17.05) 0.32+-	1.38s (1.31, 1.43) 0.04+-
master	15.45s (15.26, 15.69) 0.13+-	1.42s (1.40, 1.44) 0.02+-	13.95s (13.89, 14.03) 0.05+-	16.86ms (15.64, 17.75) 0.69+-	1.38s (1.35, 1.41) 0.02+-
evaluation	+0.53%, 0.08s invariant (0.47d, 0.40p, 0.17std)	-0.10%, -0.0s invariant (-0.05d, 0.93p, 0.03std)	+1.02%, 0.14s slower X (1.69d, 0.01p, 0.08std)	-2.43%, -0.4ms invariant (-0.74d, 0.20p, 0.51std)	+0.35%, 0.0s invariant (0.14d, 0.80p, 0.03std)

[ffreyer]

It would be good to get some feedback on how these changes feel or if there's anything else worth adding to make the 3D camera feel better.

For example I think that for orthographic cameras things are disappearing too quickly.

[ffreyer]

Refimg tests all look fine. The differences should all just be minor changes in the zoom level from center!()

[ffreyer]

At default fov the only major difference between old_cam3d! and cam3d! I can notice is that translations are too fast for the old one (around default zoom). Since the old cam doesn't scale translation speed with zoom this will get more different when changing fov. (Same with cad = true)

I think Camera3D is pretty much just a better, more feature rich version of old_cam3d! now.

Regarding tests - there are some small differences in the default zoom level I believe, which only GLMakie tests catch. So nothing to worry about.

[asinghvi17]

I just tried this on mac with touchpad, felt pretty good overall but some speeds definitely needed adjusting for trackpad.

Rotations were way to fast, and zoom was way too slow. What worked for me was:

cam = cameracontrols(ax.scene)
cam.settings.mouse_rotationspeed[] = 0.5
cam.settings.mouse_zoomspeed[] = 2

which made things a lot more responsive.

[ffreyer]

Touchpad Windows: I agree, your defaults feel better
Mouse (same) Windows: Rotation speed is ok either way I think, zoomspeed feels better at 1 for me.
Mouse Linux: I feel like rotationspeed = 1 is already on the slow side and zoomspeed = 2 is on the faster side.

It seems like mouse speed is roughly 1:2 between windows and linux for me - matching that 0.5 factor. With a scatter(rand(Point3f, 10)) I can do a 360° rotation by moving the cursor 1 window width with about 1.5 rotationspeed on linux and about 0.75 on windows. Not sure if that's a sane default to aim at, but maybe it helps normalizing things across OS's?

For zooming I think it's very dependent on trackpad vs mouse, which we can't really tell apart 🙁

[ffreyer]

Alright I went through and tested DataInspector. I fixed a bunch of small issues which were probably unrelated.

The errors in GLMakie are related to minor changes in the default camera position/zoom. The GLMakie specific tests have higher precision and notice this, the normal refiimages do not. I adjusted the camera interaction tests in Makie with new numbers so that those don't fail.

I still need to make a few tweaks to documentation. After that this will be ready to merge again.

[ffreyer]

I hard coded the camera setup here (and below) to avoid test errors. I don't know if we want to keep this or update refimgs so I'm pointing it out. The difference is tiny.

[github-actions]

Missing reference images

Found 1 new images without existing references.
Upload new reference images before merging this PR.

[ffreyer]

I'm closing this since it's part of the beta branch now