forked from Insta360-Research-Team/RomniStereo
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy patheval.py
163 lines (131 loc) · 6.04 KB
/
eval.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
# File author: Hualie Jiang ([email protected])
from __future__ import print_function, division
from argparse import ArgumentParser
import logging
import time
import multiprocessing
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
# Torch libs
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
# Internal modules
from dataset import Dataset, MultiDataset
from utils.common import *
from utils.image import *
from module.network import ROmniStereo
# Initialize
torch.backends.cudnn.benchmark = True
torch.backends.cuda.benchmark = True
parser = ArgumentParser(description='Evaluation for ROmniStereo')
parser.add_argument('--name', default='ROmniStereo', help="name of your experiment")
parser.add_argument('--restore_ckpt', help="restore checkpoint")
parser.add_argument('--db_root', default='../omnidata', type=str, help='path to dataset')
parser.add_argument('--dbname', default='omnithings', type=str,
choices=['omnithings', 'omnihouse', 'sunny', 'cloudy', 'sunset'], help='databases to evaluation')
# data options
parser.add_argument('--phi_deg', type=float, default=45.0, help='phi_deg')
parser.add_argument('--equirect_size', type=int, nargs='+', default=[160, 640], help="size of out ERP.")
parser.add_argument('--valid_iters', type=int, default=12,
help='number of flow-field updates during validation forward pass')
parser.add_argument('--vis', action='store_true', help='oneline visualization')
parser.add_argument('--save_result', action='store_true', help='save inverse depth prediction results')
parser.add_argument('--save_misc', action='store_true', help='save misc')
parser.add_argument('--save_point_cloud', action='store_true', help='save point cloud')
args = parser.parse_args()
opts = Edict()
opts.snapshot_path = args.restore_ckpt
opts.name = args.name
opts.dbname = args.dbname
opts.db_root = args.db_root
opts.data_opts = Edict()
opts.data_opts.color_aug = False
opts.data_opts.phi_deg = args.phi_deg
opts.data_opts.equirect_size = args.equirect_size
opts.valid_iters = args.valid_iters
opts.net_opts = Edict()
# Results
opts.vis = args.vis
opts.save_result = args.save_result
opts.save_misc = args.save_misc
opts.save_point_cloud = args.save_point_cloud
snapshot_name = osp.splitext(osp.basename(opts.snapshot_path))[0]
opts.result_dir = osp.join('../results', opts.dbname)
opts.out_invdepth_fmt = osp.join(opts.result_dir, 'invdepth_%05d_'+snapshot_name+'.tiff')
opts.out_misc_fmt = osp.join(opts.result_dir, 'misc_%05d_'+snapshot_name+'.png')
opts.out_point_fmt = osp.join(opts.result_dir, 'pc_%05d_'+snapshot_name+'.ply')
if opts.vis:
fig = plt.figure(frameon=False, figsize=(25, 10), dpi=40)
plt.ion()
plt.show()
def main():
if not osp.exists(opts.snapshot_path):
sys.exit('%s does not exsits' % (opts.snapshot_path))
snapshot = torch.load(opts.snapshot_path)
opts.net_opts = snapshot['net_opts']
net = torch.nn.DataParallel(ROmniStereo(opts.net_opts), device_ids=[0])
net.load_state_dict(snapshot['net_state_dict'])
opts.data_opts.use_rgb = opts.net_opts.use_rgb
opts.data_opts.num_invdepth = opts.net_opts.num_invdepth
opts.data_opts.num_downsample = opts.net_opts.num_downsample
data = Dataset(opts.dbname, opts.data_opts, db_root=opts.db_root, train=False)
grids = [torch.tensor(grid, requires_grad=False).cuda() for grid in data.grids]
if not osp.exists(opts.result_dir):
os.makedirs(opts.result_dir, exist_ok=True)
LOG_INFO('"%s" directory created' % (opts.result_dir))
eval_list = data.opts.test_idx
errors = np.zeros((len(eval_list), 5))
acc_toc = 0
for d in range(len(eval_list)):
fidx = eval_list[d]
imgs, gt, valid, raw_imgs = data.loadSample(fidx)
toc, toc2 = 0, 0
net.eval()
tic = time.time()
imgs = [torch.Tensor(img).unsqueeze(0).cuda() for img in imgs]
with torch.no_grad():
invdepth_idx = net(imgs, grids, opts.valid_iters, test_mode=True)
invdepth_idx = toNumpy(invdepth_idx[0, 0])
invdepth = data.indexToInvdepth(invdepth_idx)
toc = time.time() - tic
acc_toc += toc
# Compute errors
if len(gt) > 0: errors[d, :] = data.evalError(invdepth_idx, gt, valid)
# Visualization
if opts.vis or opts.save_misc or opts.save_point_cloud:
tic2 = time.time()
vis_img, inputs_rgb, pano_rgb, invdepth_rgb, err = data.makeVisImage(raw_imgs, invdepth, gt, return_all=True)
if opts.vis:
fig.clf()
plt.imshow(vis_img)
plt.axis('off')
plt.tight_layout()
plt.draw()
plt.pause(0.5)
if opts.save_misc:
writeImage(vis_img, opts.out_misc_fmt % fidx)
writeImage(inputs_rgb, opts.out_misc_fmt.replace('misc', 'input') % fidx)
writeImage(pano_rgb, opts.out_misc_fmt.replace('misc', 'pano') % fidx)
writeImage(invdepth_rgb, opts.out_misc_fmt.replace('misc', 'idepth') % fidx)
if err is not None:
writeImage(err, opts.out_misc_fmt.replace('misc', 'err') % fidx)
if opts.save_point_cloud:
data.writePointCloud(pano_rgb, invdepth,
opts.out_point_fmt % fidx)
toc2 = toc2 + time.time() - tic2
# Save result
if opts.save_result:
tic2 = time.time()
data.writeInvdepth(invdepth,
opts.out_invdepth_fmt % fidx)
toc2 = toc2 + time.time() - tic2
LOG_INFO('Process %d/%d, MAE: %.3f, %.3f s, misc: %.3f s' %
(d + 1, len(eval_list), errors[d, 3], toc, toc2))
mean_errors = errors.mean(axis=0)
LOG_INFO('>1: %.3f, >3: %.3f, >5: %.3f, MAE: %.3f, RMS: %.3f, Avg time: %.3f' %
(mean_errors[0], mean_errors[1], mean_errors[2], mean_errors[3], mean_errors[4], acc_toc / len(eval_list)))
if __name__ == "__main__":
main()