Sender | Message | Time |
---|---|---|
20 Jun 2023 | ||
But, definately havn't spent enough time on it, figured i'd ask here, incase I can short cut it. | 01:17:39 | |
With build123d one can do: print(Box(1, 1, 1).show_topology()) which results in:
It wouldn't be difficult to hack the code to create a dictionary. | 01:19:15 | |
How much trouble would the inverse operation be for build123d? | 01:19:39 | |
Like dict -> toposhape | 01:19:48 | |
It's work for sure but it can be done. | 01:20:12 | |
That's really helpful regardless, i'll read through it. | 01:20:13 | |
It'll be an open source project, so it'd plug right into build123 which will be cool. | 01:20:40 | |
https://build123d.readthedocs.io/en/latest/direct_api_reference.html#topology.Shape.show_topology There's a link to the source there. | 01:20:45 | |
Thanks a lot | 01:20:56 | |
What is a differentiable brep ? I'm having an hard time understanding what data you need and what you want to do with it | 05:45:42 | |
differentiable brep would mean you compute a gradient on changes to do gradient based optimisation. The data I need would be most analogous to dumping the occ (geometric and toplogical info) brep to json (a python dict). I've start making pretty good progress thanks to rogers suggestion. Should have it by the end of the day, just fixing some wires issues on the reconstruction. The data model will look something like this once I am done, with python primitives.
Its just a serialisation layer so I can pop it into gpu accelerated math functions. | 05:53:28 | |
* differentiable brep would mean you can compute a gradient on changes to do gradient based optimisation. The data I need would be most analogous to dumping the occ (geometric and toplogical info) brep to json (a python dict). I've start making pretty good progress thanks to rogers suggestion. Should have it by the end of the day, just fixing some wires issues on the reconstruction. The data model will look something like this once I am done, with python primitives.
Its just a serialisation layer so I can pop it into gpu accelerated math functions. | 05:53:49 | |
Ie geom surface would just be primitives. class GeomCylindricalSurface(GeomSurface): location: Tuple[float, float, float] normal: Tuple[float, float, float] radius: float I'll turn this into a pytorch/numpy type once I can make sure the brep->datamodel->brep works. | 05:57:36 | |
What abot brep surfaces ? It's not as simple as cylindrical surface | 06:06:35 | |
* What about brep surfaces ? It's not as simple as cylindrical surface | 06:11:23 | |
I currently convert to all of the OCC.Core.Geom types. | 06:18:34 | |
Mhhm ok curious to see what you'll end up doing. I still doesn't fully understand the goal haha | 06:49:39 | |
I suppose it all depends on you frame of reference. I'm damaged from work. Anything above 50ms is slow. I don't think I've ever been able to produce things in cad that was done in that time. | 08:40:08 | |
09:54:46 | ||
09:54:48 | ||
Yes of course, slow and fast have meaning only relative to something else. Computting a model from scratch can be slow from a human perspective but computing each operation one at a time (like in interactive GUI CAD) is most of the time instantaneous. | 10:26:15 | |
20:18:56 | ||
21 Jun 2023 | ||
09:26:22 | ||
09:32:43 | ||
10:53:22 | ||
10:53:24 | ||
13:26:42 | ||
13:26:43 | ||
14:36:42 | ||
14:36:45 |