def main():
ROOT = 'tests/meshes/bowling_Pin/'
FNAME = '10492_Bowling Pin_v1_max2011_iteration-2.obj'
mesh = loadMesh(fname=ROOT+FNAME)
mesh_xyz = mesh.vertices.T
m_norm = np.average(np.linalg.norm(mesh_xyz, axis=0))
mesh_xyz /= m_norm
norms, stdd = project(mesh_xyz)
obs = mk_obsrv(stdd, norms)
coords, weights, xyz = get_lebdev_grid()
#coords = leb2sk(coords)
interpolator = Interpolator(obs, r=1.0)
interpolated = interpolator(coords)
SHFit.theta = coords[:, 0] + np.pi
SHFit.phi = coords[:, 1]
SHFit.xyz = xyz.copy()
coefficients = []
for n in range(30):
for l in range(-n, n+1):
sph_harm_pts = make_harmonic(m=l, n=n,
theta=SHFit.theta, phi=SHFit.phi)
integral = 4*np.pi*np.sum(weights*np.real(sph_harm_pts)*interpolated[:,2])
fit = SHFit(l, n, integral)
print(fit)
coefficients.append(fit)
fig, ax = plt.subplots(nrows=1, ncols=3, sharey=False)
cs = ax[0].scatter(obs[:, 1], obs[:, 0], c=obs[:, 2])
cbar = fig.colorbar(cs)
ax[1].scatter(interpolated[:, 1], interpolated[:, 0], c=interpolated[:, 2])
ax[2].scatter(coords[:, 0], coords[:, 1], c=sph_harm_pts[:])
plt.show()
from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure()
ax1 = fig.add_subplot(121, projection='3d')
ax2 = fig.add_subplot(122, projection='3d')
#new_pts = xyz * (np.real(sph_harm_pts)**2)[:,None]
#new_pts = norms * stdd
new_pts = np.zeros(SHFit.xyz.shape, dtype=SHFit.xyz.dtype)
for c in coefficients:
new_pts += c.reconstructed
#print(new_pts)
ax1.scatter(*new_pts.T)
ax2.scatter(*(xyz*interpolated[:,2][:,None]).T, alpha=0.2)
plt.show()
def encode(target_img,
num_steps=1000,
w_avg_samples=10000,
initial_learning_rate=0.1,
initial_noise_factor=0.05,
lr_rampdown_length=0.25,
lr_rampup_length=0.05,
noise_ramp_length=0.75,
regularize_noise_weight=1e5,
device='cuda',
verbose=True):
global G
assert G is not None, 'Error: no model loaded'
# Load target image
if isinstance(target_img, np.ndarray):
target_img = PIL.Image.fromarray(target_img.astype(np.uint8))
w, h = target_img.size
s = min(w, h)
target_img = target_img.crop(
((w - s) // 2, (h - s) // 2, (w + s) // 2, (h + s) // 2))
target_img = target_img.resize((G.img_resolution, G.img_resolution),
PIL.Image.LANCZOS)
target_img = np.array(target_img, dtype=np.uint8)
target_img = torch.tensor(target_img.transpose([2, 0, 1]), device=device)
# copy model
G = copy.deepcopy(G).eval().requires_grad_(False).to(
device) # type: ignore
# run projector
projected_w_steps = projector.project(
G,
target=target_img,
num_steps=num_steps,
w_avg_samples=w_avg_samples,
initial_learning_rate=initial_learning_rate,
initial_noise_factor=initial_noise_factor,
lr_rampdown_length=lr_rampdown_length,
lr_rampup_length=lr_rampup_length,
noise_ramp_length=noise_ramp_length,
regularize_noise_weight=regularize_noise_weight,
device=device,
verbose=verbose)
# return final step
projected_w = projected_w_steps[-1].unsqueeze(0)
return projected_w
def project_img(G, fname, num_steps):
img = prepare_img_for_projection(G, fname)
projected_w_steps = project(G, target=img, num_steps=num_steps, device=get_device(), verbose=True)
w = projected_w_steps[-1].unsqueeze(0).detach().cpu().numpy()
return w