def _test_face_areas_normals_helper(self, device, dtype=torch.float32):
"""
Check the results from face_areas cuda/cpp and PyTorch implementation are
the same.
"""
meshes = self.init_meshes(10, 200, 400, device=device)
# make them leaf nodes
verts = meshes.verts_packed().detach().clone().to(dtype)
verts.requires_grad = True
faces = meshes.faces_packed().detach().clone()
# forward
areas, normals = mesh_face_areas_normals(verts, faces)
verts_torch = verts.detach().clone().to(dtype)
verts_torch.requires_grad = True
faces_torch = faces.detach().clone()
(areas_torch,
normals_torch) = TestFaceAreasNormals.face_areas_normals_python(
verts_torch, faces_torch)
self.assertClose(areas_torch, areas, atol=1e-7)
# normals get normalized by area thus sensitivity increases as areas
# in our tests can be arbitrarily small. Thus we compare normals after
# multiplying with areas
unnormals = normals * areas.view(-1, 1)
unnormals_torch = normals_torch * areas_torch.view(-1, 1)
self.assertClose(unnormals_torch, unnormals, atol=1e-6)
# backward
grad_areas = torch.rand(areas.shape, device=device, dtype=dtype)
grad_normals = torch.rand(normals.shape, device=device, dtype=dtype)
areas.backward((grad_areas, grad_normals))
grad_verts = verts.grad
areas_torch.backward((grad_areas, grad_normals))
grad_verts_torch = verts_torch.grad
self.assertClose(grad_verts_torch, grad_verts, atol=1e-6)
def face_areas_normals():
mesh_face_areas_normals(verts, faces)
torch.cuda.synchronize()
