threedle
ssue with mesh deformation using TextDeformer on Objaverse dataset
Hi, a great work on the TextDeformer project! I am attempting to use TextDeformer to deform some of my meshes from the Objaverse dataset.
in PoissonSystem.py", line 109:
return CholeskySolverD(self.n, self.inds[0,:], self.inds[1,:], self.vals, MatrixType.COO)
ValueError: Matrix is not positive definite!
I tried addressing this issue using the advice provided in https://github.com/KAIST-Visual-AI-Group/APAP/issues/3#issuecomment-2316535691, but the error persisted: Matrix is not positive definite!
To resolve this, I modified PoissonSystem.py the code as follows:
if self.my_splu is None: if isinstance(self.lap,SparseMat): if USE_CHOLESPY_CPU or USE_CHOLESPY_GPU: self.my_splu = self.lap.to_cholesky() else: self.my_splu = scipy_splu(self.lap.to_coo()) else: self.my_splu = scipy_splu(self.lap) into
if self.my_splu is None: if isinstance(self.lap,SparseMat): if USE_CHOLESPY_CPU or USE_CHOLESPY_GPU: self.my_splu = scipy_splu(self.lap.to_coo()) else: self.my_splu = scipy_splu(self.lap)
This change successfully resolved the ValueError: Matrix is not positive definite! error. However, the resulting mesh deformation is significantly degraded.
https://github.com/user-attachments/assets/ebf2f791-6d39-42e3-96fe-800b407fa5cb
Could you please advise on whether TextDeformer is suitable for deforming meshes from the Objaverse dataset? If so, any suggestions on how to improve the deformation quality would be greatly appreciated.
Thank you!
