rerun-io
Stream panel density graph should be proportional to number of instances
Say we are detecting bounding boxes or feature points on a camera stream. Sometimes there are zero, sometimes there are one, sometimes five. If we log them all as one entity we get a flatline graph:
That's because the density graph is the number of log calls (=rows in the store), and doesn't know how many instances was in each call.
It would be a lot more useful if any log call with zero instances is thin, and then get thicker the more instances there are.
Recent refactors of the data density graph should make this a lot easier to implement now.
We currently scale with the number of log calls.
This issues suggests scaling with number of instances One downside with that is that an entity with many instances (e.g. a point cloud) would then completely crowd out other entities (e.g. images).
Another option is to scale with data volume (MB).
This should probably be an option at some point.