Direct delta mush python binding#153
Conversation
|
Just realized these type checks are platform dependent issues. I will revert (830880d) that commit and leave the rest as is. Shouldn't impact rest of the branch. |
|
Some Images after the most recent fix Left (Green) is Rigid Skinning, Right (Cyan) is Direct Delta Mush Skinning
|
|
This is fantastic! Could you do a rebase against the latest master? |
* Update README.md * deprecate np.float * deprecate np.int * updated deprecated package in setup * fixed types in tests * added default dtype Co-authored-by: Teseo Schneider <teseo@bluewin.ch>
| npe_function(direct_delta_mush) | ||
| npe_doc(ds_direct_delta_mush) | ||
|
|
||
| npe_arg(v, dense_double) |
There was a problem hiding this comment.
If the question is about why direct_delta_mush.cpp (https://github.com/libigl/libigl/blob/main/include/igl/direct_delta_mush.cpp#L267) is only double, I think it's because DDM has some artifacts when using single precision.
If the question is why double for our python bindings, it is because I felt that it best reflected that DDM only works on double precision. And that work needs to be done for single precision. IMaybe I can let you pick an option here
Option 1. I can leave this as a comment in the file.
Option 2. Take in float as well and cast it.
There was a problem hiding this comment.
yeah, it looks like the upstream libigl function has Affine3d hard coded. I doubt that double precision is needed, just seems like the libigl function should be better templated. I think Option 1 is fine until libigl changes (happy to merge such a PR there).
| npe_function(direct_delta_mush_precomputation) | ||
| npe_doc(ds_direct_delta_mush_precomp) | ||
|
|
||
| npe_arg(v, dense_double) |
There was a problem hiding this comment.
same options as the question above since the reason is exactly the same.
| self.assertTrue(U.shape[0] == V.shape[0]) | ||
| self.assertTrue(U.shape[1] == 3) | ||
| self.assertTrue(U.dtype == V.dtype) | ||
| self.assertFalse(np.isnan(U).any()) |
There was a problem hiding this comment.
I am assuming you meant dtype checks for omega and specific check for U. But if you have something else in mind please let me know
kishoreVen
left a comment
kishoreVen
left a comment
There was a problem hiding this comment.
Addressed comments.
4 participants
Expose Direct Delta Mush Skinning to Python
Libigl-python currently doesn't support direct delta mush skinning. The set of commits in this branch allows for this support. Both the precomputation and delta mush skinning are exposed through the direct_delta_mush.cpp binding source and are tested in test_direct_delta_mush function of the tests/test_basic.py file.
Additionally I fixed a few tests (830880d) by pointing the tests to the correct numpy datatypes. These checks seemed to be failing the tests and I wanted to have a good starting point before adding the direct delta mush skinning test.
Platform Notes
I tested all my changes on the following configuration
OS: Ubuntu 22.04.1 LTS
Type: 64 Bit
Processor: Intel® Core™ i7-9750H CPU @ 2.60GHz × 12
Memory: 16 GB
Graphics: NVIDIA Corporation TU116M [GeForce GTX 1660 Ti Mobile]
Graphics Memory: 6 GB
Driver Version: 515.86.01
CUDA Version: 11.7