generated from eliahuhorwitz/Academic-project-page-template
-
Notifications
You must be signed in to change notification settings - Fork 0
/
render_depth.py
196 lines (163 loc) · 8.3 KB
/
render_depth.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
# This is the code for rendering depth after training.
# usage:
# python render_depth.py --config configs/test/zaragoza_bunny.txt --test_volume_size 207
import os, sys
import numpy as np
import json
import random
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from tqdm import tqdm, trange
from scipy import signal
import scipy.io
import matplotlib.pyplot as plt
from run_netf_helpers import *
import open3d as o3d
from load_nlos import *
from math import ceil
import cv2
import mcubes
from fields import *
def config_parser():
import configargparse
parser = configargparse.ArgumentParser()
parser.add_argument('--config', is_config_file=True,
help='config file path')
parser.add_argument("--nlos_file", type=str, default=None,
help='input data path')
parser.add_argument("--dataset_type", type=str, default='zaragoza256',
help='options: zaragoza256 / fk')
# NeTF arguments
parser.add_argument("--num_sampling_points", type=int, default=16,
help='number of sampling points in one direction, so the number of all sampling points is the square of this value')
parser.add_argument("--histogram_batchsize", type=int, default=1,
help='the batchsize of histogram')
parser.add_argument("--start", type=int, default=100,
help='the start point of histogram')
parser.add_argument("--end", type=int, default=300,
help='the end point of histogram')
parser.add_argument("--gt_times", type=float, default=100,
help='scaling factor of histogram')
parser.add_argument("--num_epochs", type=int, default=10,
help='number of training epochs')
parser.add_argument("--rng", type=int, default=1,
help='random seed')
# our options
parser.add_argument("--init_lr", type=float, default=1e-4,
help="initial learning rate")
parser.add_argument("--weight_h", type=float, default=1.,
help="weight for histogram loss")
parser.add_argument("--weight_ei", type=float, default=1e-1,
help="weight for eikonal loss")
parser.add_argument("--weight_z", type=float, default=0.,
help="weight for zero-sdf loss")
parser.add_argument("--weight_f", type=float, default=0.,
help="weight for freespace loss")
parser.add_argument("--weight_en", type=float, default=0.,
help="weight for entropy loss")
parser.add_argument("--save_m", type=int, default=16,
help="model save interval")
parser.add_argument("--transient_threshold", type=float, default=0,
help="threshold value for transient mask")
parser.add_argument("--geometric_init", type=int, default=1,
help="require geometric initialization (1) or not (0)")
parser.add_argument("--num_sampling_zero_pts", type=int, default=32,
help="number of sampling points on each sphere for zero-sdf loss")
parser.add_argument("--num_sampling_eik_pts", type=int, default=4096,
help="number of sampling points for eikonal loss")
parser.add_argument("--num_sampling_lb_pts", type=int, default=4096,
help="number of sampling points for freespace loss")
parser.add_argument("--render_background", type=int, default=0,
help="render background (1) or not (0)")
parser.add_argument("--test_epoch", type=int, default=5)
parser.add_argument("--test_m", type=int, default=0)
parser.add_argument("--test_volume_size", type=int, default=207)
parser.add_argument("--num_iters", type=int, default=100)
parser.add_argument("--require_mask", type=int, default=1)
parser.add_argument("--remove_mesh_th", type=float, default=0.1)
return parser
def extract_mesh():
parser = config_parser()
args = parser.parse_args()
seed = args.rng
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
create_dir("recon")
# Load data
if args.dataset_type == "zaragoza256":
nlos_data, camera_position, camera_grid_size, camera_grid_positions, camera_grid_points, volume_position, volume_size, deltaT, c = load_zaragoza256_data(args.nlos_file)
elif args.dataset_type == "fk":
nlos_data, camera_position, camera_grid_size, camera_grid_positions, camera_grid_points, volume_position, volume_size, deltaT, c = load_fk_data(args.nlos_file)
# target volume is centered at the origin
camera_grid_positions = camera_grid_positions - volume_position[:,None]
volume_position = np.zeros(3)
vmin = volume_position - volume_size / 2
vmax = volume_position + volume_size / 2
scene = os.path.splitext(os.path.basename(args.config))[0]
dir_param_list = [scene,
args.weight_h, args.weight_ei, args.weight_z, args.weight_f, args.weight_en,
args.transient_threshold,
args.num_sampling_zero_pts,
args.render_background,
]
out_dir = os.path.join("out", "_".join([str(param) for param in dir_param_list]))
# create models
sdf_network = SDFNetwork(d_in=3,
d_out=257,
d_hidden=256,
n_layers=8,
skip_in=[4],
multires=6,
bias=0.5,
scale=1.0,
geometric_init=True,
weight_norm=True,
inside_outside=False)
color_network = RenderingNetwork(d_feature=256,
mode="no_normal",
d_in=6,
d_out=1,
d_hidden=256,
n_layers=4,
weight_norm=True,
multires_view=4,
squeeze_out=False)
deviation_network = SingleVarianceNetwork(init_val=0.3)
load_model(sdf_network, "sdf", out_dir, args.test_epoch, args.test_m)
load_model(color_network, "color", out_dir, args.test_epoch, args.test_m)
load_model(deviation_network, "deviation", out_dir, args.test_epoch, args.test_m)
unit_distance = (volume_size) / (args.test_volume_size - 1)
xv = yv = zv = np.linspace(-volume_size / 2, volume_size / 2, args.test_volume_size)
coords = np.stack(np.meshgrid(xv, yv, zv),-1) # coords
coords = coords.transpose([1,0,2,3])
view_dir = np.array([0.,-1.,0.])
view_dir = torch.from_numpy(view_dir).float().unsqueeze(0).to(device)
start_pts = coords[:,-1,:]
start_pts = torch.from_numpy(start_pts.reshape(-1,3)).float().to(device)
pts_all = torch.zeros(0,3)
with torch.no_grad():
print("sphere tracing ...")
print("num iters: {}".format(args.num_iters))
curr_pts = start_pts
mask = torch.ones(curr_pts.shape[0])
for i in tqdm(range(args.num_iters)):
sdf = sdf_network(curr_pts)[:,:1]
curr_pts = curr_pts + mask[:,None] * view_dir * sdf
mask[torch.any(curr_pts< -volume_size*0.5, dim=1)] = 0
mask[torch.any(curr_pts> volume_size*0.5, dim=1)] = 0
curr_pts.requires_grad = True
_, normals = sdf_network.gradient(curr_pts, activation=None)
normals = normals.squeeze() # N_samples x 3
normals = normals / torch.norm(normals, dim=1, keepdim=True)
curr_pts = curr_pts.reshape(args.test_volume_size, args.test_volume_size, 3)
depth = np.mean(camera_grid_positions[1]) - curr_pts[:,:,1]
np.save("recon/{}_depth".format(scene), depth.to("cpu").detach().numpy())
normals = normals.reshape(args.test_volume_size, args.test_volume_size, 3)
np.save("recon/{}_normal".format(scene), normals.to("cpu").detach().numpy())
if __name__=='__main__':
torch.set_default_tensor_type('torch.cuda.FloatTensor')
extract_mesh()