RHM.torch.rhm_solver¶

Functions

`bijectivity_energy`(P12X2, P21, X2, A2, area2)	Computes the bijectivity energy
`get_energies`(mesh1, mesh2, X1, X2, P12, P21, ...)	Computes the energies
`harmonic_energy`(P12X2, W1, area2)	Computes the harmonic energy
`penalty_energy`(P12X2, X12, A1, area1, area2)	Computes the coupling energy
`rhm_refine`(mesh1, mesh2, P12_init, P21_init)	Refines a functional map using the RHM algorithm.
`rhm_refine_fast`(mesh1, mesh2, X1, X2, ...[, ...])	Solve RHM refinement with embeddings
`solve_P12`(X1, X2, X12, X21, area1, area2, ...)	Solve for the pointwise map
`solve_X12_fast`(X2, P12, P21, vertex_areas1, ...)	Solves for transferred embedding X12

RHM.torch.rhm_solver.rhm_refine(mesh1, mesh2, P12_init, P21_init, alpha=0.0005, beta=0.005, emb_dim=8, nit_max=200, nit_min=20, abs_tol=1e-09, log=False, precise=True, last_precise=True, n_jobs=1, device='cpu', verbose=False)¶

Refines a functional map using the RHM algorithm.

This solves for the optimal pointwise map between two meshes using the RHM algorithm using the RHM energy:

$E_{RHM} = \alpha E_{dirichlet} + (1-\alpha)E_{bij} + \beta E_{coupling}$$

Parameters:

mesh1 (TriMesh) – The source mesh.
mesh2 (TriMesh) – The target mesh.
P12_init (densemaps.torch.maps.PointWiseMap) – The initial map from mesh1 to mesh2. (N1, N2)
P21_init (densemaps.torch.maps.PointWiseMap) – The initial map from mesh2 to mesh1. (N2, N1)
alpha (float) – The weight of the harmonic energy term. Bijectivity is weighted by 1-alpha.
beta (float) – The weight of the penalty energy term.
emb_dim (int) – The dimension of the embedding space for vertices (distance mimics the geodesic distance).
nit_max (int) – The maximum number of iterations.
nit_min (int) – The minimum number of iterations.
abs_tol (float) – The absolute tolerance for the stopping criterion.
n_jobs (int) – The number of parallel jobs to run for NN queries
log (bool) – Whether to return the values of the energy at each step.
precise (bool) – Whether to use the precise computation for the energy terms (else, use simple nn queries).
last_precise (bool) – Whether to use the precise computation for the last iteration. Only used if precise is False. Else always True
device (str) – The device to use for the computation. ‘cpu’ or ‘cuda:{id}’. Default is ‘cpu’.
verbose (bool) – Whether to display a progress bar.

Returns:

P12 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh1 to mesh2.
P21 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh2 to mesh1.
energy_log (list, optional) – The energy values at each iteration if log is True.

RHM.torch.rhm_solver.rhm_refine_fast(mesh1, mesh2, X1, X2, P12_init, P21_init, alpha=0.0005, beta=0.005, nit_max=200, nit_min=20, abs_tol=1e-09, log=False, precise=True, last_precise=True, verbose=False)¶

Solve RHM refinement with embeddings

Parameters:

mesh1 (TriMesh) – The source mesh.
mesh2 (TriMesh) – The target mesh.
X1 (torch.Tensor) – The embedding of mesh1 vertices. (N1, p)
X2 (torch.Tensor) – The embedding of mesh2 vertices. (N1, p)
P12_init (densemaps.torch.maps.PointWiseMap) – The initial map from mesh1 to mesh2. (N1, N2)
P21_init (densemaps.torch.maps.PointWiseMap) – The initial map from mesh2 to mesh1. (N2, N1)
alpha (float) – The weight of the harmonic energy term. Bijectivity is weighted by 1-alpha.
beta (float) – The weight of the penalty energy term.
emb_dim (int) – The dimension of the embedding space for vertices (distance mimics the geodesic distance).
nit_max (int) – The maximum number of iterations.
nit_min (int) – The minimum number of iterations.
abs_tol (float) – The absolute tolerance for the stopping criterion.
n_jobs (int) – The number of parallel jobs to run for NN queries
log (bool) – Whether to return the values of the energy at each step.
precise (bool) – Whether to use the precise computation for the energy terms (else, use simple nn queries).
last_precise (bool) – Whether to use the precise computation for the last iteration. Only used if precise is False. Else always True
verbose (bool) – Whether to display a progress bar.

Returns:

P12 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh1 to mesh2.
P21 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh2 to mesh1.
energy_log (list, optional) – The energy values at each iteration if log is True.

RHM.torch.rhm_solver.solve_P12(X1, X2, X12, X21, area1, area2, alpha, beta, precise=True, faces2=None)¶

Solve for the pointwise map

Solves $(1-\alpha) E_{bij}(P_{12}, X_{21}, X_1) + \beta * E_{coupling}(P_{12}, X_2, X_{12})$

Expresses the problem in the form $min_{P_{12}}||P_{12} @ A - B||_2^2

where $A = w_{bij} * X_2$ and $B = w_{bij} * X_1 + w_{couple} * X_{12}$

Parameters:

X1 (torch.Tensor) – The embedding of mesh1 vertices. (N1, p)
X2 (torch.Tensor) – The embedding of mesh2 vertices. (N2, p)
X12 (torch.Tensor) – The embedding of mesh1 vertices in the target space. (N1, p)
X21 (torch.Tensor) – The embedding of mesh2 vertices in the source space. (N2, p)
area1 (float) – The total area of mesh1.
area2 (float) – The total area of mesh2.
alpha (float) – The weight of the harmonic energy term. Bijectivity is weighted by 1-alpha.
beta (float) – The weight of the penalty energy term.
n_jobs (int) – The number of parallel jobs to run for NN queries
precise (bool) – Whether to use the precise computation for the energy terms (else, use simple nn queries).
faces2 (torch.Tensor, optional) – The faces of mesh2 if precise is True.

Returns:

P12 – The refined map from mesh1 to mesh2. (N1, N2)

Return type:

densemaps.torch.maps.PointWiseMap

RHM.torch.rhm_solver.solve_X12_fast(X2, P12, P21, vertex_areas1, vertex_areas2, W1, alpha, beta)¶

Solves for transferred embedding X12

Solves $\alpha E_{dirichlet}(X_{12}) + (1-\alpha) E_{bij}(P_{21}, X_{12}, X_2) + \beta E_{coupling}(X_{12}, X_2, P_{12})$

Expresses the problem in the form $min_{X_{12}}||A X_{12} - B||_2^2

And solves the dual problem $A^T A X_{12} = A^T B$

Parameters:

X2 (torch.Tensor) – The embedding of mesh2 vertices. (N2, p)
P12 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh1 to mesh2. (N1, N2)
P21 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh2 to mesh1. (N2, N1)
vertex_areas1 (torch.Tensor) – The area of each vertex in mesh1. (N1,)
vertex_areas2 (torch.Tensor) – The area of each vertex in mesh2. (N2,)
W1 (sparse.csr_matrix) – The laplacian matrix of mesh1. (N1, N1)
alpha (float) – The weight of the harmonic energy term. Bijectivity is weighted by 1-alpha.
beta (float) – The weight of the penalty energy term.

Returns:

X12 – The refined embedding of mesh1 vertices in the target space. (N1, p)

Return type:

torch.Tensor

RHM.torch.rhm_solver.harmonic_energy(P12X2, W1, area2)¶

Computes the harmonic energy

$E_{dirichlet}(X_{12}) = \frac{1}{A_2} \|X_{12}\|_{W_1}$

Parameters:

P12X2 (torch.Tensor) – $P_{12} X_2$. The embedding of mesh1 vertices in the target space. (N1, p)
W1 (sparse.csr_matrix) – The laplacian matrix of mesh1. (N1, N1)
area2 (float) – The total area of mesh2.

Returns:

energy – The harmonic energy

Return type:

float

RHM.torch.rhm_solver.penalty_energy(P12X2, X12, A1, area1, area2)¶

Computes the coupling energy

$E_{coupling}(X_{12}, P_{12}X_2) = \frac{1}{A_1 A_2} \|X_{12} - P_{12}X_2\|_{A_1}^2$

Parameters:

P12X2 (torch.Tensor) – $P_{12} X_2$. The embedding of mesh1 vertices in the target space. (N1, p)
X12 (torch.Tensor) – The embedding of mesh1 vertices in the target space. (N1, p)
A1 (sparse.csr_matrix) – The area matrix of mesh1. (N1, N1)
area1 (float) – The total area of mesh1.
area2 (float) – The total area of mesh2.

Returns:

energy – The penalty energy

Return type:

float

RHM.torch.rhm_solver.bijectivity_energy(P12X2, P21, X2, A2, area2)¶

Computes the bijectivity energy

$E_{bij}(P_{21}, X_{12}, X_2) = \frac{1}{A_2} \|X_{2} - P_{21}X_{12}\|_{A_2}^2$

Parameters:

P12X2 (torch.Tensor) – $P_{12} X_2$. The embedding of mesh1 vertices in the target space. (N1, p)
P21 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh2 to mesh1. (N2, N1)
X2 (torch.Tensor) – The embedding of mesh2 vertices. (N2, p)
A2 (sparse.csr_matrix) – The area matrix of mesh2. (N2, N2)
area2 (float) – The total area of mesh2.

Returns:

energy – The bijectivity energy

Return type:

float

RHM.torch.rhm_solver.get_energies(mesh1, mesh2, X1, X2, P12, P21, X12, X21, alpha, beta)¶

Computes the energies

Parameters:

mesh1 (TriMesh) – The source mesh.
mesh2 (TriMesh) – The target mesh.
X1 (torch.Tensor) – The embedding of mesh1 vertices. (N1, p)
X2 (torch.Tensor) – The embedding of mesh2 vertices. (N2, p)
P12 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh1 to mesh2. (N1, N2)
P21 (densemaps.torch.maps.PointWiseMap) – The refined map from mesh2 to mesh1. (N2, N1)
X12 (torch.Tensor) – The embedding of mesh1 vertices in the target space. (N1, p)
X21 (torch.Tensor) – The embedding of mesh2 vertices in the source space. (N2, p)
alpha (float) – The weight of the harmonic energy term. Bijectivity is weighted by 1-alpha.
beta (float) – The weight of the penalty energy term.

Returns:

energies – The harmonic, bijectivity, coupling and total energies

Return type:

list