def get_cam_params(self, angles): K, T = camera.look_at_origin(self.fow) angles_cam = angles * (self.angle_range[1]-self.angle_range[0]) + self.angle_range[0] R = camera.euler(angles_cam) R, T = _broadcast(R, T) P = tf.matmul(camera.hom(T,'P'), camera.hom(R,'R'))[:,:3,:] return K, P
def align_bunny(xyz):
xyz_min = tf.reduce_min(xyz, axis=[0, 1], keepdims=True)
xyz_max = tf.reduce_max(xyz, axis=[0, 1], keepdims=True)
xyz = xyz - xyz_min - (xyz_max - xyz_min) / 2.
r = tf.sqrt(tf.reduce_max(tf.reduce_sum(xyz**2, -1)))
xyz /= r
angles = tf.constant([[-180., 0., 0.]]) * np.pi / 180.
R = camera.hom(camera.euler(angles), 'R')[:, :3, :]
xyz = camera.transform(R, xyz)
return xyz
def get_camera_params(field_of_view, angles):
R = camera.euler(angles)
K, T = camera.look_at_origin(field_of_view)
K, T, R = utils._broadcast(K, T, R)
P = tf.matmul(T, camera.hom(R, 'R'))
return K, P
norm = mesh.vert_normals(xyz, triangles)
lambertian_shading = tf.maximum(
tf.reduce_sum(norm * light_direction, -1, keepdims=True), 0.)
rgb_lit = rgb * (lambertian_shading + ambient)
return rgb_lit
xyz, rgb, triangles = load_bunny(filename)
light_direction = tf.constant([[-1., -0.5, -1.]])
light_direction = tf.linalg.l2_normalize(light_direction, -1)
ambient = 0.3
rgb_lit = light(xyz, rgb, triangles, (1. - ambient) * light_direction, ambient)
imsize = [256, 256]
grid_size = [1, 3]
angle_range = tf.constant([[0., -60., 0.], [0., 60., 0.]]) * np.pi / 180.
angles = utils.grid_angles(grid_size, angle_range)
R = camera.hom(camera.euler(angles), 'R')
K, T = camera.look_at_origin(5. * np.pi / 180.)
T, R = utils._broadcast(T, R)
P = tf.matmul(T, R)
ima = mesh.render(xyz, rgb_lit, triangles, K, P, imsize)
ima = utils.alpha_matte(ima, 1.0) # white background
plt.figure(1)
plt.imshow(utils.stack_images(ima[..., :-1], grid_size),
interpolation='nearest')
plt.show()