-
Notifications
You must be signed in to change notification settings - Fork 2
/
carving.py
178 lines (145 loc) · 6.12 KB
/
carving.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
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
import os
import numpy as np
import open3d as o3d
import time
import torch
from PIL import Image
from utils import *
def get_bounding_box(pts, percentile=1.0, buffer=0.1):
"""Get the bounding box of a point cloud."""
xyz1 = np.percentile(pts, percentile, axis=0)
xyz2 = np.percentile(pts, 100 - percentile, axis=0)
lwh = xyz2 - xyz1
xyz1 -= buffer * lwh
xyz2 += buffer * lwh
return xyz1, xyz2
def get_grid_points(xyz1, xyz2, grid_cell_size):
"""Get grid points."""
x1, y1, z1 = xyz1
x2, y2, z2 = xyz2
x = np.arange(x1, x2, grid_cell_size)
y = np.arange(y1, y2, grid_cell_size)
z = np.arange(z1, z2, grid_cell_size)
xx, yy, zz = np.meshgrid(x, y, z)
grid_pts = np.stack([xx, yy, zz], axis=-1).reshape(-1, 3)
return grid_pts
def project_3d_to_2d(pts, w2c, K, return_dists=False):
"""Project 3D points to 2D (nerfstudio format)."""
pts = np.array(pts)
K = np.hstack([K, np.zeros((3, 1))])
pts = np.concatenate([pts, np.ones((pts.shape[0], 1))], axis=1)
pts = np.dot(pts, w2c.T)
pts[:, [1, 2]] *= -1
if return_dists:
dists = np.linalg.norm(pts[:, :3], axis=-1)
pts = np.dot(pts, K.T)
pts_2d = pts[:, :2] / pts[:, 2:]
if return_dists:
return pts_2d, dists
return pts_2d
def project_3d_to_2d_torch(pts, w2c, K, return_dists=False):
"""Project 3D points to 2D (nerfstudio format)."""
device = pts.device
K = torch.cat([K, torch.zeros((3, 1), device=device)], 1)
pts = torch.cat([pts, torch.ones((pts.shape[0], 1), device=device)], 1)
pts = torch.matmul(pts, w2c.t())
pts[:, [1, 2]] *= -1
if return_dists:
dists = torch.norm(pts[:, :3], dim=-1)
pts = torch.matmul(pts, K.t())
pts_2d = pts[:, :2] / pts[:, 2:]
if return_dists:
return pts_2d, dists
return pts_2d
def depth_to_distance(depth, K):
"""Convert depth map to distance from camera."""
h, w = depth.shape
x, y = np.meshgrid(np.arange(w), np.arange(h))
x = x.flatten()
y = y.flatten()
depth = depth.flatten()
pts = np.stack([x, y, np.ones_like(x)], axis=1)
pts = np.dot(pts, np.linalg.inv(K).T)
pts *= depth[:, None]
dists = np.linalg.norm(pts, axis=1)
dists = dists.reshape(h, w)
return dists
def depth_to_distance_torch(depth, K):
"""Convert depth map to distance from camera."""
h, w = depth.shape
x, y = torch.meshgrid(torch.arange(w), torch.arange(h))
x = x.flatten()
y = y.flatten()
depth = depth.flatten()
pts = torch.stack([x, y, torch.ones_like(x)], dim=1).float().to(depth.device)
pts = torch.matmul(pts, torch.inverse(K).t())
pts *= depth[:, None]
dists = torch.norm(pts, dim=1)
dists = dists.reshape(h, w)
return dists
def carve_numpy(pts, masks, depths, w2cs, K, dist_thr):
n_imgs = len(masks)
for i in range(n_imgs):
h, w = masks[i].shape
pts_2d, dists = project_3d_to_2d(pts, w2cs[i], K, return_dists=True)
pts_2d = np.round(pts_2d).astype(np.int32)
pts_2d = np.clip(pts_2d, 0, [w - 1, h - 1])
observed_dists = depths[i]
is_in_mask = masks[i][pts_2d[:, 1], pts_2d[:, 0]]
is_behind_depth = dists > observed_dists[pts_2d[:, 1], pts_2d[:, 0]] - dist_thr
pts = pts[is_in_mask & is_behind_depth]
return pts
def carve_torch(pts, masks, depths, w2cs, K, dist_thr, mask_only=False):
n_imgs = len(masks)
with torch.no_grad():
mask_votes = torch.zeros(len(pts), device=pts.device, dtype=torch.int32)
depth_votes = torch.zeros(len(pts), device=pts.device, dtype=torch.int32)
for i in range(n_imgs):
h, w = masks[i].shape
pts_2d, dists = project_3d_to_2d_torch(pts, w2cs[i], K, return_dists=True)
pts_2d = torch.round(pts_2d).long().to(pts.device)
pts_2d[:, 0] = torch.clamp(pts_2d[:, 0], 0, w - 1)
pts_2d[:, 1] = torch.clamp(pts_2d[:, 1], 0, h - 1)
observed_dists = depths[i]
is_in_mask = masks[i][pts_2d[:, 1], pts_2d[:, 0]]
is_behind_depth = dists > observed_dists[pts_2d[:, 1], pts_2d[:, 0]] - dist_thr
mask_votes[is_in_mask] += 1
depth_votes[is_behind_depth] += 1
if mask_only:
pts = pts[mask_votes == n_imgs]
else:
pts = pts[(mask_votes == n_imgs) & (depth_votes == n_imgs)]
return pts
def get_carved_pts(scene_dir, grid_cell_size_ns=1/512, dist_thr_ns=0.01, verbose=False, device='cuda'):
scene_name = os.path.basename(scene_dir)
pcd_file = os.path.join(scene_dir, 'ns', 'point_cloud.ply')
dt_file = os.path.join(scene_dir, 'ns', 'dataparser_transforms.json')
t_file = os.path.join(scene_dir, 'transforms.json')
img_dir = os.path.join(scene_dir, 'images')
depth_dir = os.path.join(scene_dir, 'ns', 'renders', 'depth')
pts = load_ns_point_cloud(pcd_file, dt_file)
w2cs, K = parse_transforms_json(t_file, return_w2c=True)
ns_transform, scale = parse_dataparser_transforms_json(dt_file)
imgs, masks = load_images(img_dir, return_masks=True)
depths = load_depths(depth_dir, Ks=None)
xyz1, xyz2 = get_bounding_box(pts)
grid_cell_size = grid_cell_size_ns / scale
grid_pts = get_grid_points(xyz1, xyz2, grid_cell_size)
dist_thr = dist_thr_ns / scale
grid_pts = torch.from_numpy(grid_pts).float().to(device)
masks = [torch.from_numpy(mask).to(device) for mask in masks]
depths = [torch.from_numpy(depth).to(device) for depth in depths]
w2cs = [torch.from_numpy(w2c).float().to(device) for w2c in w2cs]
K = torch.from_numpy(K).float().to(device)
carved = carve_torch(grid_pts, masks, depths, w2cs, K, dist_thr)
if verbose:
print('scene: %s, num. surface points: %d, num. carved points: %d, scale: %.4f' %
(scene_name, len(pts), len(carved), scale))
return carved, grid_cell_size
if __name__ == '__main__':
scene_dir = '/home/azhai/n2p/data/debug/B075X4J15G_ATVPDKIKX0DER'
carved, grid_cell_size = get_carved_pts(scene_dir)
carved = carved.cpu().numpy()
pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(carved)
o3d.visualization.draw_geometries([pcd])