mkofler96 / DeepSDFStruct

80%

push

github

Add homogenization (#65)

* Add test experiment for homogenization model with training artifacts

- Introduced new experiment directory for testing homogenization.
- Added latest model, optimizer parameters, latent codes, and logs as binary files.
- Created specs.json to define network architecture and training parameters.
- Added training summary in training_summary.json.
- Implemented a test function to train the homogenization model in test_train_model.py.

* Add homogenization network and update training loss calculation

* added xlim option back to plot sdf

* Update test experiment specs and data directory for homogenization model

* Refactor homogenization model tests to use shared data directory fixture

* set default dtype to float32 in test train model

* reduced test duration by reducing sampling points

* fix retry 429 error

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Fix negative Jacobian at mesh resolution 10 by filtering tets in float64

The orientation check ran in float32 but torchfem evaluates Jacobians in
float64 (after verts.to(float64)). Tets with a tiny positive float32
volume can flip negative in float64, causing the "Negative Jacobian"
exception. Compute the final validity mask in float64 so the filter is
consistent with torchfem.

* added HF secret

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>

457 of 565 branches covered (80.88%)

Branch coverage included in aggregate %.

4073 of 5092 relevant lines covered (79.99%)

0.8 hits per line