Improve 2D heuristic fallback values for image opacity

[jprochazk]

• Can't see a depth image stacked on top of a color image #9782

Previously, only segmentation images had an opacity fallback of 0.5 when other non-segmentation images were in the same view. We now keep track of the number of images per image kind, and use that to determine what the opacity should be for all image kinds that support opacity.

The images further to the front will be transparent if there are images possibly behind them.

Assuming the draw order fallback is also used, the draw order will be (front to back):

• Segmentation
• Color
• Depth ^{(does not support opacity, so we can ignore that)}

Examples of fallback opacity:

• 2x Segmentation:
  • Segmentation (1): 1.0
  • Segmentation (2): 1.0
• 1x Segmentation + 1x Color:
  • Segmentation: 0.5
  • Color: 1.0
• 1x Color + 1x Depth:
  • Color: 0.5
  • Depth: 1.0 ^(ignored)

[github-actions]

Web viewer built successfully. If applicable, you should also test it:

• I have tested the web viewer

Result	Commit	Link	Manifest
✅	22c819d	https://rerun.io/viewer/pr/9787	+nightly +main

^{Note: This comment is updated whenever you push a commit.}

[jprochazk]

So, this is annoying. I noticed some snapshot tests were failing, and after updating them, now they're failing on CI. But looking at the differences, there really aren't any. Not sure what to do here.

[emilk]

This should probably be the same for VideoVisualizer then

[Wumpf]

lgtm!

Emil is right that we should also add this to the video visualizer just the same! Please do so before merging, thx!

The image comparision changes are strange indeed! Have you checked the diff images that were changed on ci? Ci generates a zip with before/after/diff images

Is it maybe an LFS issue (before/after not being lfs)? (doesn't look like though, it says 4 lines changed for the diff)

[jprochazk]

Have you checked the diff images that were changed on ci?

Yes. There is some difference between how it renders on my machine and CI. Is there a chance that it's different llvmpipe versions, or maybe it isn't software rendering on my end?

[Wumpf]

No change that I'm aware of, no
But yeah we don't do software rendering on dev machines right now since the setup is a bit tedious/invasive and depends on platform circumstances

I'll have a poke later today!

[jprochazk]

Emil is right that we should also add this to the video visualizer just the same! Please do so before merging, thx!

It looks like video doesn't support Opacity. At a glance it seems like it could, but it feels out of scope for this PR

[jprochazk]

Opened an issue for that and linked it in a comment:

• Support Opacity for VideoFrameReference #9832

[jprochazk]

Decided to revert the snapshot to what's on CI for the time being

[jprochazk]

which is proving to be a mistake, because now I'll have to wait for CI to report all the sequential failures one by one...

[Wumpf]

Opened an issue for that and linked it in a comment:

* [Support `Opacity` for `VideoFrameReference` #9832](https://github.com/rerun-io/rerun/issues/9832)

thanks!

[jprochazk]

don't merge this yet, i left some temporary code in tests so i could get more failures in parallel

(it was supposed to be temporary, but i think it's fine to leave it in)

[github-actions]

Latest documentation preview deployed successfully.

Result	Commit	Link
✅	22c819d	https://landing-e8m7hrx5k-rerun.vercel.app/docs

^{Note: This comment is updated whenever you push a commit.}

[jprochazk]

It seems like what you pushed in b914586 didn't fix the failures 🤔

(my fault, forgot to invert the condition)