Varied edge radius for CurveNetwork

[lvjonok]

Good day!

I have been looking for a solution to render varied radius of edges with high performance and this library has been on my list for a while. I have seen, that the functionality for the varied radius has been introduced here #180, but there was no follow up about the varied edge radius since there is no simple solution since there the nodes can be of different radii.

In my particular case, I want to take advantage of varied radius of edge to display the change in extrusion for 3D printer slicer (bigger radius will mean more extrusion in particular straight line movement), and the radii of nodes are not really important for me. Also, I don't expect a situation where one node is shared by more than two edges, so it is much easier case.

Do you think this feature will be a good addition to the polyscope? If so, I would love to develop the feature given some guidance on the procedure and some useful references.

Best Regards,
Kozlov Lev

[Megan-Soo]

Hi Kozlov, I was also looking for a way to visualise networks with different edge radii. FWIW, a workaround is to generate a surface mesh based on the nodes, edges, and radii for edge tips and edge tails. Here is an example I made. I started with a 1D tree with about 60k edges. It renders quite smoothly.

[lvjonok]

Hi Kozlov, I was also looking for a way to visualise networks with different edge radii. FWIW, a workaround is to generate a surface mesh based on the nodes, edges, and radii for edge tips and edge tails. Here is an example I made. I started with a 1D tree with about 60k edges. It renders quite smoothly.

That looks really great! Am I right that I can look into how it was implemented in this repository: https://github.com/Megan-Soo/ps_demo ?

[Megan-Soo]

Yes, have a look at visualise_tree_mesh.py. The data isn't available in the repo, but some randomly generated nodes, edges, and per-edge radii will do. The main function is generate_network_tube_mesh(), which provides vertices and faces for ps.register_surface_mesh(). Hope this helps, let me know if you have any questions about it.

[nmwsharp]

Hi all, thanks for the discussion here, I love it!

I see there's significant interest in variable edges. I took a look at the code and fortunately it wasn't too difficult to add, so I jsut put together a quick commit adding support: #351. Give it a try!

I also added these functions to the Python bindings, and created a documentation page, although those won't be visible until I release the next version (hopefully very soon).

[nmwsharp]

I'll go ahead and close this for now in the hopes that it's solved, but feel free to continue discussing or re-open if it is not solved.

[Megan-Soo]

Thanks very much @nmwsharp! The new function is great, definitely a feature we're hoping for the curve network. Just one thing, the nodes are getting assigned radius values even though a scalar quantity wasn't added to them. Below is an example. Other than that, the model looks great, we're looking forward to this feature in the new version.

[nmwsharp]

Love the image!

nodes are getting assigned radius values even though a scalar quantity wasn't added to them

I went with a default policy that if you don't specify a node radius, the nodes take a radius value which is the average of the incident edge radii. I thiiiink this is a reasonable default, you can always create a separate node quantity if you want a different node radius.... but I'm open to other opinions if you feel different.

Looking at your image though, I wonder if there might be a bug in that logic, it looks as if some of the nodes are larger than any of the adjacent edges.... I'll have to take a look.

Also, would it be okay with you if I use that image in the docs or elsewhere? It's a great example of the CurveNetwork. If not no worries, but if so just let me know what name/link to give for credit. (and maybe if you could render it with a white or transparent background that would help too)

[Megan-Soo]

would it be okay with you if I use that image in the docs or elsewhere? It's a great example of the CurveNetwork. If not no worries, but if so just let me know what name/link to give for credit

Sure thing, this was generated using a volume-filling algorithm developed by Tawhai et al. (2004).

the nodes take a radius value which is the average of the incident edge radii. I thiiiink this is a reasonable default

Yup, I agree. It might not be the case at the moment though, appreciate you taking a look at it again.

create a separate node quantity if you want a different node radius

I tried assigning random values or just zeroes to the nodes - the node colours do change accordingly but the node radius don't.

Happy to provide another screenshot with transparent background when this is fixed. Thank you!

[nmwsharp]

a bug in that logic

Yep, there was a bug, using the wrong normalization factor scale edge radius vs. node. Just pushed a fix in 258cd7b. This would lead to lots of confusing behavior, so perhaps it's what you were seeing?

Otherwise, everything I can think to check using node + edge seems to be working. The demo app also includes a few lines testing curve network radii, which can be tried by passing an .obj files as input if you're interested to try that one there. Please let me know if anything still seems off!

[Megan-Soo]

It's beautiful, thank you. Pictures as promised, let me know if you want more. I will cite Polyscope in my work as well, of course.

Edit: I found out that casting the edge radii onto the nodes is sufficient to visualise the varied edge radius (see third pic).
tree.add_scalar_quantity("radius node",joint_radii,defined_on='nodes',enabled=True) tree.set_node_radius_quantity("radius node")

[nmwsharp]

Amazing, glad you pointed out the issue to track down the bug! Thanks all. Happy visualizing :)