pos_gt = torch.stack((x_gt, y_gt, z_gt), dim=-1)
imgs_gt = []
print("Rendering GT images...")
for i in trange(numsteps):
_vertices = vertices_gt.clone() + pos_gt[i]
rgba = renderer.forward(_vertices, faces_gt, textures)
imgs_gt.append(rgba)
logdir = Path(args.logdir) / args.expid
if args.log:
logdir.mkdir(exist_ok=True)
if args.log:
write_imglist_to_gif(imgs_gt,
logdir / "gt.gif",
imgformat="rgba",
verbose=False)
if args.save_timelapse:
timelapse = kaolin.visualize.Timelapse(args.logdir)
# Estimate the length of the pendulum by backprop.
parameters = []
if args.optimize_length:
length_est = torch.nn.Parameter(torch.tensor([0.5], device=device),
requires_grad=False)
lengthmodel = SimpleModel(length_est).to(device)
parameters += list(lengthmodel.parameters())
if args.optimize_gravity:
) + 1e-5
# print("method:", args.method, initial_velocity)
# if args.method == "gradsim":
# loss = (
# torch.nn.functional.mse_loss(imgs[-1], target_image)
# + torch.nn.functional.mse_loss(imgs[-2], target_image)
# + torch.nn.functional.mse_loss(imgs[-3], target_image)
# + torch.nn.functional.mse_loss(imgs[-4], target_image)
# + torch.nn.functional.mse_loss(imgs[-5], target_image)
# )
if args.log:
write_imglist_to_gif(
imgs,
os.path.join(logdir, f"{e:02d}.gif"),
imgformat="rgba",
verbose=False,
)
write_imglist_to_dir(
imgs, os.path.join(logdir, f"{e:02d}"), imgformat="rgba",
)
imageio.imwrite(
os.path.join(logdir, f"last_frame_{e:02d}.png"),
(imgs[-1][0].permute(1, 2, 0).detach().cpu().numpy() * 255).astype(
np.uint8
),
)
write_meshes_to_file(
vertices,
faces.detach().cpu().numpy(),
os.path.join(logdir, f"vertices_{e:05d}")
# sys.exit(0)
# imgs_gt.append(rgb)
# SoftRas
rgba = renderer.forward(
state.q.unsqueeze(0).to(device),
faces.unsqueeze(0).to(device),
textures.to(device),
)
imgs_gt.append(rgba)
cloth_path = Path("cache/cloth")
cloth_path.mkdir(exist_ok=True)
write_imglist_to_gif(imgs_gt,
cloth_path / "gt.gif",
imgformat="rgba",
verbose=False)
init_inv_mass = particle_inv_mass_gt.clone()
init_inv_mass[-1] = 0.0
init_inv_mass[-2] = 0.0
massmodel = SimpleModel(
# particle_inv_mass_gt + 50 * torch.rand_like(particle_inv_mass_gt),
init_inv_mass,
activation=torch.nn.functional.relu,
)
velocitymodel = SimpleModel(-0.01 * torch.ones_like(particle_velocity_gt),
activation=None)
epochs = 50
save_gif_every = 1
compare_every = 1
# print(state.rigid_x.shape)
# v_in = torch.from_numpy(np.asarray(model_gt.shape_geo_src[0].vertices)).float()
# print(torch.allclose(v_in, vertices))
vertices_current = get_world_vertices(
vertices, state_gt.rigid_r.view(-1), state_gt.rigid_x)
rgba = renderer.forward(
vertices_current.unsqueeze(0).to(device),
faces.unsqueeze(0).to(device),
textures.to(device),
)
imgs_gt.append(rgba)
positions_gt.append(state_gt.rigid_x)
logvertices_gt.append(vertices_current.detach().cpu().numpy())
if args.log:
write_imglist_to_gif(imgs_gt, os.path.join(logdir, "gt.gif"))
write_meshes_to_file(logvertices_gt,
faces.detach().cpu().numpy(),
os.path.join(logdir, "vertices_gt"))
# """
# Optimize for physical parameters.
# """
# # ke, kd, kf, mu
# shape_material_guesses = (9500, 950, 950, 0.5)
# builder = df.sim.ModelBuilder()
# rigid = builder.add_rigid_body(
# pos=pos_gt, rot=rot_gt, vel=vel_gt, omega=omega_gt,
# )
colors_bxpx3=textures_gt,
# uv_bxpx2=uv_gt,
# texture_bx3xthxtw=textureimg_gt,
# lightparam=lightparam,
)
rgba = torch.cat((img_gt, alpha_gt), dim=-1)
# rgba = renderer.forward(
# body_gt.get_world_vertices().unsqueeze(0), faces_gt, textures_gt
# )
imgs_gt.append(rgba)
# writer.append_data((255 * img).astype(np.uint8))
# writer.close()
if args.log:
write_imglist_to_gif(imgs_gt,
os.path.join(logdir, "gt.gif"),
imgformat="dibr")
# Load the template mesh (usually a sphere).
mesh = TriangleMesh.from_obj(args.template)
vertices = meshutils.normalize_vertices(
mesh.vertices.unsqueeze(0)).to(device)
faces = mesh.faces.to(device).unsqueeze(0)
# uv = get_spherical_coords_x(vertices[0].cpu().numpy())
# uv = torch.from_numpy(uv).to(device).float().unsqueeze(0) / 255.0
# Paint the mesh yellow
textures = torch.stack(
(
torch.ones(
1, vertices.shape[-2], dtype=torch.float32, device=device),
torch.ones(
)
imgs.append(rgba)
if not args.inference:
loss.backward()
optimizer.step()
optimizer.zero_grad()
tqdm.write(f"Loss: {loss.item():.5}")
render_time += 1
if args.inference:
filename = os.path.join("cache", "jellyfish", "debug", "inference")
else:
filename = os.path.join("cache", "jellyfish", "debug", f"{render_time:02d}")
write_imglist_to_gif(imgs, f"{filename}.gif", imgformat="rgba", verbose=False)
if args.inference:
filename = os.path.join("cache", "jellyfish", "debug", "inference")
else:
filename = os.path.join(
"cache", "jellyfish", "debug", f"last_frame_{render_time:02d}"
)
imageio.imwrite(
f"{filename}.png",
(imgs[-1][0].permute(1, 2, 0).detach().cpu().numpy() * 255).astype(
np.uint8
),
)
torch.save(network, "cache/jellyfish/debug/model.pt")