search

rerun-io / rerun

Visualize streams of multimodal data. Fast, easy to use, and simple to integrate. Built in Rust using egui.
https://rerun.io/
Apache License 2.0
5.8k stars 263 forks source link

Improve the orbit eye origin #5451

Open abey79 opened 4 months ago

abey79 commented 4 months ago

One aspect that I consistently found to behave is surprising way is the orbit origin of the 3D view eye, and I have reasons to believe that I'm not alone.

After spending some time in OnShape, I noticed that it uses the bbox center of the geometry visible on screen as default orbit point. Also, if an hovered object is (right-)clicked, then it is used as orbit point instead. We should consider implementing something similar.

https://github.com/rerun-io/rerun/assets/49431240/45d5b463-4f7e-4030-9753-4dd31f85b664

https://github.com/rerun-io/rerun/assets/49431240/c6d50db4-72d8-4224-a40a-821640c83ed4

abey79 commented 4 months ago

Using the centre of bounding box of all displayed objects (on screen or otherwise) would already be an improvement IMO.

abey79 commented 4 months ago

Dual cube OnShape playground: https://cad.onshape.com/documents/6183b6e53b67ad1a89aa7abb/w/cbc3cec32eb338f09732aa38/e/702de136749c7ed77764f756?renderMode=0&uiState=65eebcf38b06243496523313

Wumpf commented 4 months ago

I noticed that it uses the bbox center of the geometry visible on screen as default orbit point.

doesn't seem to be the case, once two boxes are on screen it doesn't orbit around the center of those two, only if there's a single one on screen. I'm a bit unclear what it orbits around actually.

Taking our orbit camera model as-is today and change the orbit center also means changing the camera's position, thus completely changing the screen's content. This clearly doesn't happen in onshape, they decouple this somehow further.

What might be related is that the pan operation changes the camera's center point on the image plane, not the camera position. Something we don't support at all.

Using the centre of bounding box of all displayed objects (on screen or otherwise) would already be an improvement IMO.

I have a lot of doubts about that because of the above (we can't change the camera orbit without changing the screen contents a lot) and because in many of our example scenes the bounding box jumps erratic because of outlier points.

Wumpf commented 4 months ago

someone in this thread called this style of orbiting "mcad style", maybe that gives us some pointers to look deeper into this later :)