-
Notifications
You must be signed in to change notification settings - Fork 0
/
example_usage.py
61 lines (59 loc) · 4.22 KB
/
example_usage.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
# from pygsp.graphs.nngraphs.spherehealpix import SphereHealpix
# grid_object_sphere = SphereHealpix(nside=32, n_neighbors=8)
# grid_object_sphere = SphereHealpix(subdivisions=32, k=8, nest=True)
# pygsp has many different versions.... API is different
# See https://github.com/deepsphere/deepsphere-pytorch/blob/master/notebooks/demo_visualizations.ipynb for visualization
# for k in range(len(stuff)):
# cur_stuff = stuff[k]
# data_exp = torch_geometric.data.Data()
# data_exp.radial_depth_loc = torch.from_numpy(cur_stuff["depth_xyz"][0]).to(device="cuda", non_blocking=True,
# dtype=torch.float32)
# data_exp.radial_depth_idx_tri = torch.from_numpy(cur_stuff["depth_tri"][0]).to(device="cuda", non_blocking=True,
# dtype=torch.long)
# data_exp.radial_depth_idx_ray = torch.from_numpy(cur_stuff["depth_ray"][0]).to(device="cuda", non_blocking=True,
# dtype=torch.long)
#
# data_exp.ortho_depth_loc = torch.from_numpy(cur_stuff["depth_xyz"][1]).to(device="cuda", non_blocking=True,
# dtype=torch.float32)
# data_exp.ortho_depth_idx_tri = torch.from_numpy(cur_stuff["depth_tri"][1]).to(device="cuda", non_blocking=True,
# dtype=torch.long)
# data_exp.ortho_depth_idx_ray = torch.from_numpy(cur_stuff["depth_ray"][1]).to(device="cuda", non_blocking=True,
# dtype=torch.long)
#
# data_exp.radial_prob_loc = torch.from_numpy(np.concatenate(cur_stuff["prob_xyz_radial"], axis=0)).to(device="cuda",
# non_blocking=True,
# dtype=torch.float32)
# data_exp.radial_prob_idx_tri = torch.from_numpy(np.concatenate(cur_stuff["prob_tri_radial"], axis=0)).to(
# device="cuda", non_blocking=True, dtype=torch.long)
# data_exp.radial_prob_idx_ray = torch.from_numpy(np.concatenate(cur_stuff["prob_ray_radial"], axis=0)).to(
# device="cuda", non_blocking=True, dtype=torch.long)
# data_exp.radial_offsets = torch.LongTensor([0] + [len(_) for _ in cur_stuff["prob_ray_radial"]])
#
# data_exp.ortho_prob_loc = torch.from_numpy(np.concatenate(cur_stuff["prob_xyz_ortho"], axis=0)).to(device="cuda",
# non_blocking=True,
# dtype=torch.float32)
# data_exp.ortho_prob_idx_tri = torch.from_numpy(np.concatenate(cur_stuff["prob_tri_ortho"], axis=0)).to(
# device="cuda", non_blocking=True, dtype=torch.long)
# data_exp.ortho_prob_idx_ray = torch.from_numpy(np.concatenate(cur_stuff["prob_ray_ortho"], axis=0)).to(
# device="cuda", non_blocking=True, dtype=torch.long)
# data_exp.ortho_offsets = torch.LongTensor([0] + [len(_) for _ in cur_stuff["prob_ray_ortho"]])
# data_exp.verts_in = torch.from_numpy(verts).to(device="cuda", non_blocking=True, dtype=torch.float32)
# data_exp.tri_in = torch.from_numpy(faces).to(device="cuda", non_blocking=True, dtype=torch.long)
#
# data_exp.num_nodes = 100
# data_list.append(data_exp)
# blending_obj = diff_render_blend(sgrid_obj=torch.from_numpy(grid_obj).float())
# blending_obj.cuda()
# blending_obj = torch_geometric.nn.DataParallel(blending_obj, device_ids=[0, 1])
# torch.cuda.synchronize()
# t0 = time.time()
# for k in range(10):
# output = blending_obj(data_list)
# torch.cuda.synchronize()
# print(time.time() - t0, "TIME")
# val = output[0][2]
# val = val.detach().cpu().numpy()
# # plt.scatter(x=grid_obj2.lon, y=-grid_obj2.lat, c=val, cmap="Greys", alpha=0.5)
# plt.scatter(x=grid_obj2[0], y=grid_obj2[1], c=val, cmap="jet", alpha=0.5)
# plt.colorbar()
# plt.show()