def test_compare_coarse_cpu_vs_cuda(self):
torch.manual_seed(231)
N = 1
image_size = 512
blur_radius = 0.0
bin_size = 32
max_faces_per_bin = 20
device = torch.device("cpu")
meshes = ico_sphere(2, device)
faces = meshes.faces_packed()
verts = meshes.verts_packed()
faces_verts = verts[faces]
num_faces_per_mesh = meshes.num_faces_per_mesh()
mesh_to_face_first_idx = meshes.mesh_to_faces_packed_first_idx()
args = (
faces_verts,
mesh_to_face_first_idx,
num_faces_per_mesh,
image_size,
blur_radius,
bin_size,
max_faces_per_bin,
)
bin_faces_cpu = _C._rasterize_meshes_coarse(*args)
device = torch.device("cuda:0")
meshes = ico_sphere(2, device)
faces = meshes.faces_packed()
verts = meshes.verts_packed()
faces_verts = verts[faces]
num_faces_per_mesh = meshes.num_faces_per_mesh()
mesh_to_face_first_idx = meshes.mesh_to_faces_packed_first_idx()
args = (
faces_verts,
mesh_to_face_first_idx,
num_faces_per_mesh,
image_size,
blur_radius,
bin_size,
max_faces_per_bin,
)
bin_faces_cuda = _C._rasterize_meshes_coarse(*args)
# Bin faces might not be the same: CUDA version might write them in
# any order. But if we sort the non-(-1) elements of the CUDA output
# then they should be the same.
for n in range(N):
for by in range(bin_faces_cpu.shape[1]):
for bx in range(bin_faces_cpu.shape[2]):
K = (bin_faces_cuda[n, by, bx] != -1).sum().item()
idxs_cpu = bin_faces_cpu[n, by, bx].tolist()
idxs_cuda = bin_faces_cuda[n, by, bx].tolist()
idxs_cuda[:K] = sorted(idxs_cuda[:K])
self.assertEqual(idxs_cpu, idxs_cuda)
def _test_coarse_rasterize(self, device):
image_size = 16
blur_radius = 0.2**2
bin_size = 8
max_faces_per_bin = 3
# fmt: off
verts = torch.tensor(
[
[-0.5, 0.0, 0.1], # noqa: E241, E201
[0.0, 0.6, 0.1], # noqa: E241, E201
[0.5, 0.0, 0.1], # noqa: E241, E201
[-0.3, 0.0, 0.4], # noqa: E241, E201
[0.3, 0.5, 0.4], # noqa: E241, E201
[0.75, 0.0, 0.4], # noqa: E241, E201
[-0.4, -0.3, 0.9], # noqa: E241, E201
[0.2, -0.7, 0.9], # noqa: E241, E201
[0.4, -0.3, 0.9], # noqa: E241, E201
[-0.4, 0.0, -1.5], # noqa: E241, E201
[0.6, 0.6, -1.5], # noqa: E241, E201
[0.8, 0.0, -1.5], # noqa: E241, E201
],
device=device,
)
faces = torch.tensor(
[
[1, 0, 2], # noqa: E241, E201 bin 00 and bin 01
[4, 3, 5], # noqa: E241, E201 bin 00 and bin 01
[7, 6, 8], # noqa: E241, E201 bin 10 and bin 11
[10, 9, 11
], # noqa: E241, E201 negative z, should not appear.
],
dtype=torch.int64,
device=device,
)
# fmt: on
meshes = Meshes(verts=[verts], faces=[faces])
faces_verts = verts[faces]
num_faces_per_mesh = meshes.num_faces_per_mesh()
mesh_to_face_first_idx = meshes.mesh_to_faces_packed_first_idx()
bin_faces_expected = (torch.ones(
(1, 2, 2, max_faces_per_bin), dtype=torch.int32, device=device) *
-1)
bin_faces_expected[0, 0, 0, 0:2] = torch.tensor([0, 1])
bin_faces_expected[0, 0, 1, 0:2] = torch.tensor([0, 1])
bin_faces_expected[0, 1, 0, 0:3] = torch.tensor([0, 1, 2])
bin_faces_expected[0, 1, 1, 0:3] = torch.tensor([0, 1, 2])
bin_faces = _C._rasterize_meshes_coarse(
faces_verts,
mesh_to_face_first_idx,
num_faces_per_mesh,
image_size,
blur_radius,
bin_size,
max_faces_per_bin,
)
bin_faces_same = (bin_faces.squeeze().flip(
dims=[0]) == bin_faces_expected).all()
self.assertTrue(bin_faces_same.item() == 1)
def _test_coarse_rasterize(self, device):
image_size = 16
# No blurring. This test checks that the XY directions are
# correctly oriented.
blur_radius = 0.0
bin_size = 8
max_faces_per_bin = 3
# fmt: off
verts = torch.tensor(
[
[-0.5, 0.1, 0.1], # noqa: E241, E201
[-0.3, 0.6, 0.1], # noqa: E241, E201
[-0.1, 0.1, 0.1], # noqa: E241, E201
[-0.3, -0.1, 0.4], # noqa: E241, E201
[ 0.3, 0.5, 0.4], # noqa: E241, E201
[0.75, -0.1, 0.4], # noqa: E241, E201
[ 0.2, -0.3, 0.9], # noqa: E241, E201
[ 0.3, -0.7, 0.9], # noqa: E241, E201
[ 0.6, -0.3, 0.9], # noqa: E241, E201
[-0.4, 0.0, -1.5], # noqa: E241, E201
[ 0.6, 0.6, -1.5], # noqa: E241, E201
[ 0.8, 0.0, -1.5], # noqa: E241, E201
],
device=device,
)
# Expected faces using axes convention +Y down, + X right, +Z in
# Non symmetrical triangles i.e face 0 and 3 are in one bin only
faces = torch.tensor(
[
[ 1, 0, 2], # noqa: E241, E201 bin 01 only
[ 4, 3, 5], # noqa: E241, E201 all bins
[ 7, 6, 8], # noqa: E241, E201 bin 10 only
[10, 9, 11], # noqa: E241, E201 negative z, should not appear.
],
dtype=torch.int64,
device=device,
)
# fmt: on
meshes = Meshes(verts=[verts], faces=[faces])
faces_verts = verts[faces]
num_faces_per_mesh = meshes.num_faces_per_mesh()
mesh_to_face_first_idx = meshes.mesh_to_faces_packed_first_idx()
# Expected faces using axes convention +Y down, + X right, + Z in
bin_faces_expected = (
torch.ones(
(1, 2, 2, max_faces_per_bin), dtype=torch.int32, device=device
)
* -1
)
bin_faces_expected[0, 0, 0, 0] = torch.tensor([1])
bin_faces_expected[0, 1, 0, 0:2] = torch.tensor([1, 2])
bin_faces_expected[0, 0, 1, 0:2] = torch.tensor([0, 1])
bin_faces_expected[0, 1, 1, 0] = torch.tensor([1])
# +Y up, +X left, +Z in
bin_faces = _C._rasterize_meshes_coarse(
faces_verts,
mesh_to_face_first_idx,
num_faces_per_mesh,
image_size,
blur_radius,
bin_size,
max_faces_per_bin,
)
# Flip x and y axis of output before comparing to expected
bin_faces_same = (bin_faces.squeeze() == bin_faces_expected).all()
self.assertTrue(bin_faces_same.item() == 1)
