def test_vertex_normals_random(self):
"""Tests the computation of vertex normals for a regular octahedral."""
tensor_vertex_size = np.random.randint(1, 3)
tensor_out_shape = np.random.randint(1, 5, size=tensor_vertex_size)
tensor_out_shape = tensor_out_shape.tolist()
with self.subTest(name="triangular_faces"):
vertex_on_axes = np.array(
((0., 0., 1), (1., 0., 0.), (0., 1., 0.), (0., 0., -1.),
(-1., 0., 0.), (0., -1., 0.)),
dtype=np.float32)
vertex_on_axes = vertex_on_axes.reshape([1] * tensor_vertex_size +
[6, 3])
index_init = np.array(((0, 1, 2), (0, 2, 4), (0, 4, 5), (0, 5, 1),
(3, 2, 1), (3, 4, 2), (3, 5, 4), (3, 1, 5)),
dtype=np.int32)
index_init = index_init.reshape([1] * tensor_vertex_size + [8, 3])
index_init = np.tile(index_init, tensor_out_shape + [1, 1])
vertex_scale = np.random.uniform(0.5, 5.,
tensor_out_shape + [1] * 2)
vertex_init = vertex_on_axes * vertex_scale
expected = vertex_on_axes * (vertex_scale * 0. + 1.)
vertex_tensor = tf.convert_to_tensor(value=vertex_init)
index_tensor = tf.convert_to_tensor(value=index_init)
self.assertAllClose(
normals.vertex_normals(vertex_tensor, index_tensor), expected)
with self.subTest(name="polygon_faces"):
num_vertices = np.random.randint(4, 8)
poly_vertices = []
rad_step = np.pi * 2. / num_vertices
for i in range(num_vertices):
poly_vertices.append(
[np.cos(i * rad_step),
np.sin(i * rad_step), 0])
vertex_init = np.array(poly_vertices, dtype=np.float32)
vertex_init = vertex_init.reshape([1] * tensor_vertex_size +
[-1, 3])
vertex_init = vertex_init * vertex_scale
index_init = np.arange(num_vertices, dtype=np.int32)
index_init = index_init.reshape([1] * tensor_vertex_size + [1, -1])
index_init = np.tile(index_init, tensor_out_shape + [1, 1])
expected = np.array((0., 0., 1.), dtype=np.float32)
expected = expected.reshape([1] * tensor_vertex_size + [1, 3])
expected = np.tile(expected, tensor_out_shape + [num_vertices, 1])
vertex_tensor = tf.convert_to_tensor(value=vertex_init)
index_tensor = tf.convert_to_tensor(value=index_init)
self.assertAllClose(
normals.vertex_normals(vertex_tensor, index_tensor), expected)
def test_vertex_normals_jacobian_random(self):
"""Test the Jacobian of the vertex normals function."""
tensor_vertex_size = np.random.randint(1, 3)
tensor_out_shape = np.random.randint(1, 5, size=tensor_vertex_size)
tensor_out_shape = tensor_out_shape.tolist()
vertex_axis = np.array(((0., 0., 1), (1., 0., 0.), (0., 1., 0.),
(0., 0., -1.), (-1., 0., 0.), (0., -1., 0.)),
dtype=np.float32)
vertex_axis = vertex_axis.reshape([1] * tensor_vertex_size + [6, 3])
faces = np.array(((0, 1, 2), (0, 2, 4), (0, 4, 5), (0, 5, 1),
(3, 2, 1), (3, 4, 2), (3, 5, 4), (3, 1, 5)),
dtype=np.int32)
faces = faces.reshape([1] * tensor_vertex_size + [8, 3])
index_init = np.tile(faces, tensor_out_shape + [1, 1])
vertex_scale = np.random.uniform(0.5, 5., tensor_out_shape + [1] * 2)
vertex_init = vertex_axis * vertex_scale
vertex_tensor = tf.convert_to_tensor(value=vertex_init)
index_tensor = tf.convert_to_tensor(value=index_init)
y = normals.vertex_normals(vertex_tensor, index_tensor)
self.assert_jacobian_is_correct(vertex_tensor, vertex_init, y)
def vertex_normals(vertex_tensor): return normals.vertex_normals(vertex_tensor, index_tensor)