Is there any way to use different .stl files for different geometries during the training process and as a result have a network that can accept some discrete value that represents each geometry as an input? I have seen all the tutorials in the Modulus documentation but they either use a geometry built from CSG primitives only or a gPC approach without the parameterization.
I don’t know for sure as I haven’t messed with it myself yet, but this is something I’m interested in as well. Looking at tesselation.py leads me to believe it’s possible via the parameterization input. How that’s used and what it’s used for I’m not certain on. I find the documentation and search to be difficult to navigate and find what I’m looking for, but here’s the option as seen in the source code.
https://docs.nvidia.com/deeplearning/modulus/modulus-v2209/api/modulus.geometry.html#modulus.geometry.parameterization.Parameterization
def from_stl(
cls,
filename,
airtight=True,
parameterization=Parameterization(),
):
"""
makes mesh from STL file
Parameters
----------
filename : str
filename of mesh.
airtight : bool
If the geometry is airtight or not. If false sample everywhere for interior.
parameterization : Parameterization
Parameterization of geometry.
"""
Hi @gorpinich4
Have a look at the parameterized tessellated geometry example here. If you can represent the different STL files as some sort of parameterization (that would perhaps be an input to a neural network) this should be possible.
Hey @npstrike
I find the documentation and search to be difficult to navigate and find what I’m looking for.
Thanks for the feedback, this is very useful. We are presently evaluating the documentation (we’ve just recently moved to this new system) to look for parts where we can improve. If you have any particular pain points (e.g. search doesn’t work well, structure is cumbersome) or any ideas for improvements we are always open to input from our users. Thanks!
My apologies for being non-constructively critical, I appreciate your professionalism. There is one thing in particular that I seem to be struggling with regularly. I know there are documents that cover and explain the various classes and components of Modulus (e.g. tessellation.py), and I tend to look for those under the Modulus API portion of the side bar.
For example, if I’m looking for what loss values are available for the hydra configurator out of the box, I may find my way to this file via clicking through the Modulus API docs,
https://docs.nvidia.com/deeplearning/modulus/modulus-v2209/api/modulus.loss.html
However, what I’m really looking for is this, which is more difficult for me to navigate to, part because the search function seems to apply to all nvidia products (until the modulus filter is applied) and even then doesn’t seem to always get me to this page.
https://docs.nvidia.com/deeplearning/modulus/modulus-v2209/user_guide/features/configuration.html
As I’ve been writing this response, I’ve edited this last section many times. It took me a minute or two to find the second link, but I think I’ve identified my source of confusion. The documentation I’m often looking for appears to be under “Modulus Features” rather than “Modulus API.” To me, ‘Features’ makes it sound like it would contain marketing material rather than technical use documentation and so I don’t naturally check there. As a user, I don’t fully understand that title, or how it’s contents are intended to be different from the “Modulus API” section, but I think if the relevant docs were cross linked between the two different sources, or the naming was tweaked it may be more intuitive to me.
When I skim the side bar I categorize the sections under two categories: topics intended to help me learn the platform, and secondly, reference docs (i.e. Modulus API). Perhaps this was a bad assumption on my part.
No worries, this is very useful for us. Thank you. I agree the search currently leaves much to be desired. I’ll re-initiate talks with our docs team to see if it can get fixed. I’ll forward the rest of these comments to our team for review.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.