rerun
rerun copied to clipboard
rerun-io
Configurable ranges / (spatial) extent in image and tensor view
I was fiddling around with rerun and see large potential to use it for data analysis in our field where I would like to replace matplotlib by rerun. Selecting dimensions to slice through is pretty awesome!
To actually use it (i.e. extracting coordinates intuitively, slicing through the actual dimension) I would like to have an option to add an extent (or axis limits) similar to matplotlib: https://matplotlib.org/stable/tutorials/intermediate/imshow_extent.html
The option to add axes and a grid would help as well.
Note: the data I'm visualizing here is 3D (X, Y, Z in meter) with some quality.
It is related to #683
Hey @tfachmann! Thanks for opening this issue and really fantastic to hear you see a large potential in Rerun!
It's almost funny that this issue came in during this weekend because on Friday, we aligned internally around that during the next 6 weeks, Rerun should become a completely valid and significantly better alternative to imshow. The first week or two will be spent mostly on laying the ground work so I can't promise when this particular issue will be solved, but at least know it's a top priority for us right now.
Cool, nice to hear it aligns with what you're trying tackle anyway! Looking forward for upcoming releases and can't wait to show it in our institute 🚀
So what you are asking for is basically a cropping of the tensor, correct?
Do you want/need this to be precise, or do you really just want to zoom in to the tensor?
In my case, not really (it may be that it is not 100% extent what I'm looking for). I basically want to give real world coordinates to my data (per axis). E.g. I know that my X axis starts at -1.5 meter and ends at 1.2 meter. For hovering / slicing and other operations I would expect rerun to show the respective values of the given Dimension instead of simple ranges e.g. 0-250 that I must calculate back manually
Ah I see - you want to map the tensor indices to some other coordinate system. This could perhaps be handled by the transform system then.
Thanks for the clarification!
I recently revisted the TensorView and ran into this use-case again. Now, after time have passed, is there any work in this direction or an existing solution to map tensor indices to some other coordinate system?
Unfortunately we didn't really find time to work on this—which admittedly is somewhat sad considering how long this request has been open.
