-
Notifications
You must be signed in to change notification settings - Fork 5
/
surface_reconstruction.py
241 lines (202 loc) · 12 KB
/
surface_reconstruction.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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
'''
Author: Xingtong Liu, Maia Stiber, Jindan Huang, Masaru Ishii, Gregory D. Hager, Russell H. Taylor, and Mathias Unberath
Copyright (C) 2020 Johns Hopkins University - All Rights Reserved
You may use, distribute and modify this code under the
terms of the GNU GENERAL PUBLIC LICENSE Version 3 license for non-commercial usage.
You should have received a copy of the GNU GENERAL PUBLIC LICENSE Version 3 license with
this file. If not, please write to: [email protected] or [email protected]
'''
import cv2
import numpy as np
from pathlib import Path
import argparse
import h5py
import tqdm
# Local
import cg_utils
import models
def display_depth_map(depth_map, min_value=None, max_value=None, colormode=cv2.COLORMAP_JET, scale=None):
if (min_value is None or max_value is None) and scale is None:
if len(depth_map[depth_map > 0]) > 0:
min_value = np.min(depth_map[depth_map > 0])
else:
min_value = 0.0
elif scale is not None:
min_value = 0.0
max_value = scale
else:
pass
depth_map_visualize = np.abs((depth_map - min_value) / (max_value - min_value + 1.0e-8) * 255)
depth_map_visualize[depth_map_visualize > 255] = 255
depth_map_visualize[depth_map_visualize <= 0.0] = 0
depth_map_visualize = cv2.applyColorMap(np.uint8(depth_map_visualize), colormode)
return depth_map_visualize
def surface_mesh_global_scale(surface_mesh):
max_bound = np.max(surface_mesh.vertices, axis=0)
min_bound = np.min(surface_mesh.vertices, axis=0)
return np.linalg.norm(max_bound - min_bound, ord=2), np.linalg.norm(min_bound, ord=2), np.abs(
max_bound[2] - min_bound[0])
def main():
cv2.destroyAllWindows()
parser = argparse.ArgumentParser(
description='Depth fusion and surface reconstruction',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--data_root', type=str, required=True,
help='root storing the data')
parser.add_argument('--visualize_fused_model', action='store_true',
help='whether or not to visualize fused 3d model')
parser.add_argument('--trunc_margin_multiplier', type=float, default=10.0,
help='truncate margin factor of the signed distance function')
parser.add_argument('--sequence_root', type=str, default=None,
help='root of one video sequence')
parser.add_argument('--patient_id', nargs='+', type=int, required=True,
help='list of patient id')
parser.add_argument('--max_voxel_count', type=float, default=400.0 ** 3,
help='maximum count of voxels for depth map fusion')
args = parser.parse_args()
folder_list = list(Path(args.data_root).rglob("_start*"))
folder_list.sort()
for patient_id in args.patient_id:
data_root = Path(args.data_root) / "{}".format(patient_id)
sub_folders = list(data_root.glob("_start*/"))
sub_folders.sort()
for folder in sub_folders:
print("Processing {}...".format(str(folder)))
if args.sequence_root is not None:
if str(folder) != args.sequence_root:
continue
# Read hdf5 file
hdf5_path = folder / "fusion_data.hdf5"
if not hdf5_path.exists():
print("{} not exists".format(str(hdf5_path)))
continue
fusion_data = h5py.File(str(hdf5_path), 'r', libver='latest')
print("Estimating voxel volume bounds...")
vol_bnds = np.zeros((3, 2))
fusion_data_mask_array = fusion_data["mask"]
_, height, width, _ = fusion_data_mask_array.shape
mask_boundary = fusion_data_mask_array[0].astype(np.float32).reshape((height, width, 1))
fusion_data_mean_depth = fusion_data["mean_depth"]
fusion_data_std_depth = fusion_data["std_depth"]
fusion_data_extrinsics = fusion_data["extrinsics"]
fusion_data_intrinsics = fusion_data["intrinsics"]
fusion_data_color = fusion_data["color"]
n_imgs = fusion_data_mean_depth.shape[0]
cam_pose_list = []
cam_intr = fusion_data_intrinsics[0, :, :].reshape((3, 3))
for i in range(n_imgs):
mean_depth_im = fusion_data_mean_depth[i, :, :, :].reshape((height, width, 1))
cam_pose_list.append(fusion_data_extrinsics[i, :, :].reshape((4, 4)))
# Compute camera view frustum and extend convex hull
view_frust_pts = models.get_view_frustum(mean_depth_im, cam_intr, cam_pose_list[i])
vol_bnds[:, 0] = np.minimum(vol_bnds[:, 0], np.amin(view_frust_pts, axis=1))
vol_bnds[:, 1] = np.maximum(vol_bnds[:, 1], np.amax(view_frust_pts, axis=1))
# Avoid nan sample
if np.any(np.isnan(np.asarray(cam_pose_list))):
print("NAN sequence encountered in {}".format(folder))
continue
voxel_size = 0.1
vol_dim = (vol_bnds[:, 1] - vol_bnds[:, 0]) / voxel_size
# Adaptively change the size of one voxel to fit into the GPU memory
volume = vol_dim[0] * vol_dim[1] * vol_dim[2]
factor = (volume / args.max_voxel_count) ** (1.0 / 3.0)
voxel_size *= factor
print("voxel size: {}".format(voxel_size))
# Initialize voxel volume
print("Initializing voxel volume...")
tsdf_vol = models.TSDFVolume(vol_bnds, voxel_size=voxel_size,
trunc_margin=voxel_size * args.trunc_margin_multiplier)
# Loop through images and fuse them together
print("Integrating depth images...")
overall_mean_depth_value = 0.0
color_image_list = []
for i in range(n_imgs):
mean_depth_im = fusion_data_mean_depth[i, :, :, :].reshape((height, width, 1))
mean_depth_im = mean_depth_im * mask_boundary
if i == 0:
overall_mean_depth_value = np.sum(mean_depth_im) / np.sum(mask_boundary).astype(np.float32)
else:
overall_mean_depth_value = overall_mean_depth_value * (i / (i + 1.0)) + \
np.sum(mean_depth_im) / np.sum(mask_boundary) * (1 / (i + 1.0))
tq = tqdm.tqdm(total=n_imgs)
tq.set_description('Depth fusion')
for i in range(n_imgs):
# Read RGB-D image and camera pose
color_image = fusion_data_color[i, :, :, :].reshape((height, width, 3))
color_image_list.append(color_image)
mean_depth_im = fusion_data_mean_depth[i, :, :, :].reshape((height, width, 1))
mean_depth_im = mean_depth_im * mask_boundary
std_depth_im = fusion_data_std_depth[i, :, :, :].reshape((height, width, 1))
std_depth_im = std_depth_im * mask_boundary
cam_pose = cam_pose_list[i]
# Integrate observation into voxel volume (assume color aligned with depth)
# We have changed the slope of the truncated distance function based on the depth std values
tsdf_vol.integrate(color_image, mean_depth_im, cam_intr, cam_pose,
min_depth=1.0e-3 * overall_mean_depth_value,
std_im=std_depth_im,
obs_weight=1.)
tq.update(1)
tq.close()
verts, faces, norms, colors, _ = tsdf_vol.get_mesh(only_visited=True)
fused_model_path = str(folder / "fused_mesh.ply")
print("Writing mesh model...")
models.meshwrite(fused_model_path, verts, faces, -norms, colors)
print("Loading scene model...")
scene, surface_mesh = cg_utils.load_3d_model(fused_model_path)
mesh_global_scale, min_distance, z_distance = surface_mesh_global_scale(surface_mesh)
print("Mesh global scale: {}...".format(mesh_global_scale))
rendering_color_image_list = []
depth_image_list = []
print("Fly-through rendering...")
tq = tqdm.tqdm(total=n_imgs)
tq.set_description('Rendering')
for i in range(n_imgs):
# 4x4 rigid transformation matrix T^(world)_(camera)
cam_pose = cam_pose_list[i]
rendering_color_image, depth_image = cg_utils.get_depth_image_from_3d_model(scene, cam_intr,
height, width, cam_pose,
z_near=1.0e-3 * overall_mean_depth_value,
z_far=mesh_global_scale,
point_light_strength=3.0 * voxel_size,
ambient_strength=0.5)
rendering_color_image = cv2.cvtColor(rendering_color_image, cv2.COLOR_RGBA2RGB)
rendering_color_image_list.append(
(rendering_color_image * mask_boundary).reshape((height, width, 3)).astype(np.uint8))
mask_image = (depth_image > 0.0).astype(np.float32)
depth_image_list.append(mask_image * depth_image)
tq.update(1)
max_depth = np.max(np.asarray(depth_image_list))
max_std_depth = np.max(fusion_data_std_depth)
GIF_image_list = []
for i, simulated_depth_image in enumerate(depth_image_list):
display_simulated_depth_image = display_depth_map(depth_map=simulated_depth_image * mask_boundary.
reshape((height, width)),
colormode=cv2.COLORMAP_JET, scale=max_depth)
display_simulated_depth_image = cv2.cvtColor(display_simulated_depth_image, cv2.COLOR_BGR2RGB)
predicted_mean_depth_image = fusion_data_mean_depth[i, :, :, :].reshape((height, width, 1))
display_predicted_depth_image = display_depth_map(depth_map=predicted_mean_depth_image,
colormode=cv2.COLORMAP_JET, scale=max_depth)
display_predicted_depth_image = cv2.cvtColor(display_predicted_depth_image, cv2.COLOR_BGR2RGB)
predicted_std_depth_image = fusion_data_std_depth[i, :, :, :].reshape((height, width, 1))
display_predicted_std_image = display_depth_map(predicted_std_depth_image, min_value=0.0,
max_value=max_std_depth,
colormode=cv2.COLORMAP_HOT)
display_predicted_std_image = cv2.cvtColor(display_predicted_std_image, cv2.COLOR_BGR2RGB)
GIF_image_list.append(cv2.hconcat(
[color_image_list[i], rendering_color_image_list[i], display_simulated_depth_image,
display_predicted_depth_image, display_predicted_std_image]))
if args.visualize_fused_model:
cv2.imshow("video_rendering_depth", cv2.cvtColor(GIF_image_list[i], cv2.COLOR_RGB2BGR))
cv2.waitKey(10)
print("Writing fly-through video of fused mesh...")
result_video_fp = cv2.VideoWriter(
str(folder / "fused_mesh.avi"),
cv2.VideoWriter_fourcc(*'DIVX'), 20,
(GIF_image_list[0].shape[1], GIF_image_list[0].shape[0]))
for i in range(len(GIF_image_list)):
result_video_fp.write(cv2.cvtColor(GIF_image_list[i], cv2.COLOR_RGB2BGR))
result_video_fp.release()
if args.visualize_fused_model:
cv2.destroyAllWindows()
if __name__ == '__main__':
main()