-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconvert-to-standard-ply.py
191 lines (170 loc) · 8.75 KB
/
convert-to-standard-ply.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
179
180
181
182
183
184
185
186
187
188
189
190
from plyfile import PlyData, PlyElement
import numpy as np
from io import BytesIO
import torch
from torchsummary import summary
class Octree:
def __init__(self, min_coord, max_coord, max_depth):
self.min_coord = min_coord
self.max_coord = max_coord
self.max_depth = max_depth
self.children = [None] * 8
self.indices = []
def insert(self, point, index, depth=0):
if depth == self.max_depth:
self.indices.append(index)
return
octant = self.get_octant(point)
if self.children[octant] is None:
mid_coord = (self.min_coord + self.max_coord) / 2
child_min = np.where(octant & (1 << np.arange(3)), mid_coord, self.min_coord)
child_max = np.where(octant & (1 << np.arange(3)), self.max_coord, mid_coord)
self.children[octant] = Octree(child_min, child_max, self.max_depth)
self.children[octant].insert(point, index, depth + 1)
def get_octant(self, point):
mid_coord = (self.min_coord + self.max_coord) / 2
octant = 0
for i in range(3):
if point[i] >= mid_coord[i]:
octant |= 1 << i
return octant
def get_sorted_indices(self, indices):
for child in self.children:
if child is not None:
child.get_sorted_indices(indices)
indices.extend(self.indices)
def convert_ply(SourceDir, OutputDir, ply_file_path, camera_position):
plydata = PlyData.read(SourceDir + ply_file_path)
vert = plydata["vertex"]
min_coord = np.array([vert["x"].min(), vert["y"].min(), vert["z"].min()])
max_coord = np.array([vert["x"].max(), vert["y"].max(), vert["z"].max()])
octree = Octree(min_coord, max_coord, 5)
for i in range(len(vert)):
point = np.array([vert["x"][i], vert["y"][i], vert["z"][i]])
octree.insert(point, i)
sorted_indices = []
octree.get_sorted_indices(sorted_indices)
anchor_pos = []
gsplat_positions = []
anchor_scales = []
anchor_features = []
anchor_levels = []
anchor_extra_levels = []
anchor_info = []
offset_names = [p.name for p in vert.properties if p.name.startswith("f_offset")]
offset_names = sorted(offset_names, key=lambda x: int(x.split('_')[-1]))
offsets = np.zeros((len(vert["x"]), len(offset_names)))
for idx, attr_name in enumerate(offset_names):
offsets[:, idx] = np.asarray(vert[attr_name]).astype(np.float32)
offsets = offsets.reshape((offsets.shape[0], 3, -1))
for i in sorted_indices:
anchor_pos.append(vert["x"][i])
anchor_pos.append(vert["y"][i])
anchor_pos.append(vert["z"][i])
anchor_levels.append(vert["level"][i])
anchor_extra_levels.append(vert["extra_level"][i])
for j in range(10):
gsplat_positions.append(vert["x"][i] + offsets[i][0][j] * np.exp(vert["scale_0"][i]))
gsplat_positions.append(vert["y"][i] + offsets[i][1][j] * np.exp(vert["scale_1"][i]))
gsplat_positions.append(vert["z"][i] + offsets[i][2][j] * np.exp(vert["scale_2"][i]))
for j in range(3):
anchor_scales.append(np.exp(vert["scale_" + str(j + 3)][i]))
for j in range(32):
anchor_features.append(vert["f_anchor_feat_" + str(j)][i])
gaussians_num_per_anchor = 10
anchor_num = len(anchor_pos) // 3
gaussians_num = anchor_num * gaussians_num_per_anchor
dtype = [('x', 'f4'), ('y', 'f4'), ('z', 'f4'),
('nx', 'f4'), ('ny', 'f4'), ('nz', 'f4'),
('f_dc_0', 'f4'), ('f_dc_1', 'f4'), ('f_dc_2', 'f4'),
('f_rest_0', 'f4'), ('f_rest_1', 'f4'), ('f_rest_2', 'f4'), ('f_rest_3', 'f4'), ('f_rest_4', 'f4'),
('f_rest_5', 'f4'), ('f_rest_6', 'f4'), ('f_rest_7', 'f4'), ('f_rest_8', 'f4'), ('f_rest_9', 'f4'),
('f_rest_10', 'f4'), ('f_rest_11', 'f4'), ('f_rest_12', 'f4'), ('f_rest_13', 'f4'), ('f_rest_14', 'f4'),
('f_rest_15', 'f4'), ('f_rest_16', 'f4'), ('f_rest_17', 'f4'), ('f_rest_18', 'f4'), ('f_rest_19', 'f4'),
('f_rest_20', 'f4'), ('f_rest_21', 'f4'), ('f_rest_22', 'f4'), ('f_rest_23', 'f4'), ('f_rest_24', 'f4'),
('f_rest_25', 'f4'), ('f_rest_26', 'f4'), ('f_rest_27', 'f4'), ('f_rest_28', 'f4'), ('f_rest_29', 'f4'),
('f_rest_30', 'f4'), ('f_rest_31', 'f4'), ('f_rest_32', 'f4'), ('f_rest_33', 'f4'), ('f_rest_34', 'f4'),
('f_rest_35', 'f4'), ('f_rest_36', 'f4'), ('f_rest_37', 'f4'), ('f_rest_38', 'f4'), ('f_rest_39', 'f4'),
('f_rest_40', 'f4'), ('f_rest_41', 'f4'), ('f_rest_42', 'f4'), ('f_rest_43', 'f4'), ('f_rest_44', 'f4'),
('opacity', 'f4'),
('scale_0', 'f4'), ('scale_1', 'f4'), ('scale_2', 'f4'),
('rot_0', 'f4'), ('rot_1', 'f4'), ('rot_2', 'f4'), ('rot_3', 'f4')]
gaussian_data = np.zeros(gaussians_num, dtype=dtype)
gaussian_data['x'] = gsplat_positions[::3][:gaussians_num]
gaussian_data['y'] = gsplat_positions[1::3][:gaussians_num]
gaussian_data['z'] = gsplat_positions[2::3][:gaussians_num]
color_mlp = torch.jit.load( SourceDir + "color_mlp.pt")
cov_mlp = torch.jit.load( SourceDir + "cov_mlp.pt")
opacity_mlp = torch.jit.load( SourceDir + "opacity_mlp.pt")
anchor_features_np = np.array(anchor_features)
anchor_pos_np = np.array(anchor_pos)
camera_position_np = np.array(camera_position)
anchor_feature_with_camera = np.zeros((anchor_num, 35), dtype=np.float32)
anchor_feature_with_camera[:, :32] = anchor_features_np.reshape(anchor_num, 32)
dx = anchor_pos_np[::3] - camera_position_np[0]
dy = anchor_pos_np[1::3] - camera_position_np[1]
dz = anchor_pos_np[2::3] + camera_position_np[2]
distance = np.sqrt(dx**2 + dy**2 + dz**2)
anchor_feature_with_camera[:, 32] = dx / distance
anchor_feature_with_camera[:, 33] = dy / distance
anchor_feature_with_camera[:, 34] = dz / distance
anchor_feature_with_camera_tensor = torch.from_numpy(anchor_feature_with_camera).unsqueeze(1).unsqueeze(1)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
anchor_feature_with_camera_tensor = anchor_feature_with_camera_tensor.to(device)
with torch.no_grad():
color_output = color_mlp(anchor_feature_with_camera_tensor)
cov_output = cov_mlp(anchor_feature_with_camera_tensor)
opacity_output = opacity_mlp(anchor_feature_with_camera_tensor)
color_output = color_output.squeeze().cpu().numpy()
color_output = color_output.reshape(-1)
cov_output = cov_output.squeeze().cpu().numpy()
cov_output = cov_output.reshape(-1)
opacity_output = opacity_output.squeeze().cpu().numpy()
opacity_output = opacity_output.reshape(-1)
SH_C0 = 0.28209479177387814
gaussian_data['f_dc_0'] = (color_output[::3][:gaussians_num] - 0.5) / SH_C0
gaussian_data['f_dc_1'] = (color_output[1::3][:gaussians_num] - 0.5) / SH_C0
gaussian_data['f_dc_2'] = (color_output[2::3][:gaussians_num] - 0.5) / SH_C0
for i in range(gaussians_num):
if opacity_output[i] < 0:
gaussian_data['opacity'][i] = -100
else:
gaussian_data['opacity'][i] = - np.log(1/opacity_output[i] - 1)
for i in range(gaussians_num):
gaussian_data['scale_0'][i] = np.log(sigmoid(cov_output[i * 7 + 0]) * anchor_scales[(i//gaussians_num_per_anchor) * 3 + 0])
gaussian_data['scale_1'][i] = np.log(sigmoid(cov_output[i * 7 + 1]) * anchor_scales[(i//gaussians_num_per_anchor) * 3 + 1])
gaussian_data['scale_2'][i] = np.log(sigmoid(cov_output[i * 7 + 2]) * anchor_scales[(i//gaussians_num_per_anchor) * 3 + 2])
cov_vec_0 = []
cov_vec_1 = []
cov_vec_2 = []
cov_vec_3 = []
for i in range(gaussians_num):
vec = cov_output[i * 7 + 3:i * 7 + 7]
vec = vec / np.linalg.norm(vec)
cov_vec_0.append(vec[0])
cov_vec_1.append(vec[1])
cov_vec_2.append(vec[2])
cov_vec_3.append(vec[3])
gaussian_data['rot_0'] = cov_vec_0
gaussian_data['rot_1'] = cov_vec_1
gaussian_data['rot_2'] = cov_vec_2
gaussian_data['rot_3'] = cov_vec_3
opacity_threshold = -10
filtered_gaussian_data = []
for i in range(gaussians_num):
if gaussian_data['opacity'][i] > opacity_threshold:
gaussian_data_item = tuple(gaussian_data[i])
filtered_gaussian_data.append(gaussian_data_item)
vertex_element = PlyElement.describe(np.array(filtered_gaussian_data, dtype=dtype), 'vertex')
ply_data = PlyData([vertex_element])
ply_data.write('output.ply')
plydata = PlyData.read('output.ply')
print(f"Number of vertices in output.ply: {len(plydata['vertex'])}")
print("Conversion complete.")
def sigmoid(x):
return 1 / (1 + np.exp(-x))
if __name__ == "__main__":
SourceDir = "./gs-data/"
OutputDir = "./converted/"
camera_position = np.array([0, 0, 0])
convert_ply(SourceDir, OutputDir, "point_cloud.ply", camera_position)