-
Notifications
You must be signed in to change notification settings - Fork 1
/
eval.py
277 lines (216 loc) · 11 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
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
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
import os
import time
import math
import argparse
import skimage.io
import os.path as osp
from PIL import Image
import matplotlib.pyplot as plt
import torch.nn as nn
import torch.nn.functional as F
from torchvision import transforms
import torchvision.transforms.functional as TF
from utils import *
from models.metric import *
from dataloader import readpfm as rp
import dataloader.middleburyinferlist as listM
import dataloader.sceneflowinferlist as s_lst
# Argparse Load Saved Model
parser = argparse.ArgumentParser()
parser.add_argument('--cuda', default='0', type=str, help='Select GPU. Default: 0')
parser.add_argument('--loadmodel', default=None, help='Path to state_dict. Default: None')
parser.add_argument('--model', default='PSMNet', help='Select network: [PSMNet/ GwcNet /CFNet]. Default: PSMNet')
parser.add_argument('--maxdisp', type=int, default=192, help='Maximum disparity range. Default: 192')
parser.add_argument('--savepath', default=None, help='Path to directory for saving disparity maps. Default: None')
# Dataset
## KITTI
parser.add_argument('--kitti15', action='store_true', default=False, help='Test using KITTI2015 dataset. Default: False')
parser.add_argument('--kitti12', action='store_true', default=False, help='Test using KITTI2012 dataset. Default: False')
## Middlebury
parser.add_argument('--midFull', action='store_true', default=False, help='Test using Middlebury-Full dataset. Default: False')
parser.add_argument('--midHalf', action='store_true', default=False, help='Test using Middlebury-Half dataset. Default: False')
parser.add_argument('--midQuar', action='store_true', default=False, help='Test using Middlebury-Quarter dataset. Default: False')
## ETH3D
parser.add_argument('--eth', action='store_true', default=False, help='Test using ETH3D dataset. Default: False')
parser.add_argument('--no_cuda', action='store_true', default=False, help='Disable CUDA training. Default: False')
parser.add_argument('--seed', type=int, default=1, metavar='S', help='Random seed (Default: 1)')
parser.add_argument('--verbose', action='store_true', default=False, help='Print progress for each sample. Default: False')
args = parser.parse_args()
kitti = args.kitti12 or args.kitti15
mid = args.midFull or args.midHalf or args.midQuar
torch.manual_seed(args.seed)
np.random.seed(args.seed)
if not args.no_cuda:
torch.cuda.manual_seed(args.seed)
os.environ["CUDA_VISIBLE_DEVICES"] = args.cuda
print('>>>[INFO] Using GPU: {}'.format(args.cuda))
else:
print('>>>[INFO] Not using GPU')
if args.model == 'PSMNet':
from models.PSMNet.stackhourglass import PSMNet
model = PSMNet(args.maxdisp)
elif args.model == 'GwcNet':
from models.GwcNet.stackhourglass import GwcNet_G
model = GwcNet_G(args.maxdisp)
elif args.model == 'CFNet':
from models.CFNet.stackhourglass import CFNet
model = CFNet(args.maxdisp)
else:
raise Exception("Invalid network is selected. Expected one from [PSMNet, GwcNet, CFNet].")
model = nn.DataParallel(model)
model.cuda()
# Load testing data
if args.kitti15:
print(">>>[INFO] Loading data: KITTI 2015")
filepath = '/media/SSD2/wei/Dataset/KITTI/data_scene_flow/train/'
test_left, test_right, test_disp = dataloader(filepath, submission=False)
if args.savepath:
save_path = osp.join(args.savepath, 'KITTI', '2015')
elif args.kitti12:
print(">>>[INFO] Loading data: KITTI 2012")
filepath = '/media/SSD2/wei/Dataset/KITTI/data_stereo_flow/training/'
test_left, test_right, test_disp = dataloader(filepath, submission=False)
if args.savepath:
save_path = osp.join(args.savepath, 'KITTI', '2012')
elif args.midFull:
print(">>>[INFO] Loading data: Middlebury Full")
filepath = '/media/SSD2/wei/Dataset/Middlebury-Full/'
[test_left, test_right, test_disp, mask_val] = listM.dataloader(filepath)
if args.savepath:
save_path = osp.join(args.savepath, 'Middlebury', 'Full')
elif args.midHalf:
print(">>>[INFO] Loading data: Middlebury Half")
filepath = '/media/SSD2/wei/Dataset/Middlebury-Half/'
[test_left, test_right, test_disp, mask_val] = listM.dataloader(filepath)
if args.savepath:
save_path = osp.join(args.savepath, 'Middlebury', 'Half')
elif args.midQuar:
print(">>>[INFO] Loading data: Middlebury Quarter")
filepath = '/media/SSD2/wei/Dataset/Middlebury-Quarter/'
[test_left, test_right, test_disp, mask_val] = listM.dataloader(filepath)
if args.savepath:
save_path = osp.join(args.savepath, 'Middlebury', 'Quarter')
elif args.eth:
print(">>>[INFO] Loading data: ETH3D")
filepath = '/media/SSD2/wei/Dataset/ETH3D/'
[test_left, test_right, test_disp, mask_val] = listM.dataloader(filepath)
if args.savepath:
save_path = osp.join(args.savepath, 'eth')
else: # Scene Flow
print(">>>[INFO] Loading data: FlyingThings3D")
filepath = '/media/SSD2/wei/SceneFlow/'
[test_left, test_right, test_disp] = s_lst.dataloader(filepath)
if args.savepath:
save_path = osp.join(args.savepath, 'SceneFlow')
if args.savepath is not None:
print('>>>[INFO] Saving to {}'.format(save_path))
else:
print('>>>[INFO] Not saving outputs')
state_dict_list = []
if os.path.isdir(args.loadmodel): # if a directory with checkpoints is selected
state_dict_list = os.listdir(args.loadmodel)
else: # path to a specific checkpoint (e.g. ckpt.tar)
state_dict_list.append(args.loadmodel)
print('>>>[INFO] Found {} ckpts in {}'.format(len(state_dict_list), args.loadmodel))
# Loop through ckpt_list
for i in range(len(state_dict_list)):
mean_d1 = 0
mean_epe = 0
avg_time = 0
if len(state_dict_list) > 1:
ckpt_path = os.path.join(args.loadmodel, 'ckpt_' + str(i+1) + '.tar')
else:
ckpt_path = state_dict_list[i]
print(">>>[INFO] Loading: {}".format(ckpt_path))
model = load_ckpt(model, state_dict_path=ckpt_path)
for idx in range(len(test_left)):
left_o = Image.open(test_left[idx]).convert('RGB')
right_o = Image.open(test_right[idx]).convert('RGB')
if kitti:
disp_o = Image.open(test_disp[idx])
disp_true = np.ascontiguousarray(disp_o, dtype=np.float32)
disp_true = torch.tensor(disp_true, dtype=torch.float32)
disp_true = disp_true / 256
elif (mid or args.eth):
mask_o = Image.open(mask_val[idx])
disp_o, _ = disparity_loader(test_disp[idx])
disp_true = np.ascontiguousarray(disp_o, dtype=np.float32)
disp_true = torch.tensor(disp_true, dtype=torch.float32)
mask_o = torch.from_numpy(np.ascontiguousarray(mask_o, dtype=np.float32))
mask = (disp_true > 0) & (disp_true < args.maxdisp) & (mask_o == 255)
else: #SceneFlow
disp_o, _ = disparity_loader(test_disp[idx])
disp_true = np.ascontiguousarray(disp_o, dtype=np.float32)
disp_true = torch.tensor(disp_true, dtype=torch.float32)
left = process(left_o).numpy()
right = process(right_o).numpy()
left = np.reshape(left, [1, 3, left.shape[1], left.shape[2]])
right = np.reshape(right, [1, 3, right.shape[1], right.shape[2]])
top_padded_size = math.ceil(left.shape[2] / 64) * 64
left_padded_size = math.ceil(left.shape[3] / 64) * 64
top_pad = top_padded_size - left.shape[2]
left_pad = left_padded_size - left.shape[3]
left = np.lib.pad(left, ((0, 0), (0, 0), (top_pad, 0), (0, left_pad)), mode='constant', constant_values=0)
right = np.lib.pad(right, ((0, 0), (0, 0), (top_pad, 0), (0, left_pad)), mode='constant', constant_values=0)
left, right = torch.from_numpy(left), torch.from_numpy(right)
start_time = time.time()
model.eval()
with torch.no_grad():
dispPred = model(left.cuda(), right.cuda())
dispPred = dispPred.data.cpu()
dispPred = torch.squeeze(dispPred)
time_per_img = time.time() - start_time
avg_time += time_per_img
# Remove padding (if any)
if top_pad > 0:
if left_pad == 0:
dispPred = dispPred[top_pad:, :]
else:
dispPred = dispPred[top_pad:, :-left_pad]
else:
if left_pad > 0:
dispPred = dispPred[:, :-left_pad]
# Performance evaluation
if kitti:
# D1 (3-px)
epe_err = epe_metric(dispPred, disp_true, (disp_true > 0)&(disp_true < args.maxdisp))
d1_err = d1_metric(dispPred, disp_true, (disp_true > 0)&(disp_true < args.maxdisp))
elif mid:
# 2-px
epe_err = epe_metric(dispPred, disp_true, mask)
d1_err = thres_metric(dispPred, disp_true, thres=2, mask=mask)
imgName = test_left[idx].split('/')[-2]
if imgName in ['PianoL', 'Playroom', 'Playtable', 'Shelves', 'Vintage']:
d1_err = d1_err * 0.5
elif args.eth:
# 1-px
epe_err = epe_metric(dispPred, disp_true, mask)
d1_err = thres_metric(dispPred, disp_true, thres=1, mask=mask)
else:
epe_err = epe_metric(dispPred, disp_true, (disp_true < args.maxdisp) & (disp_true > 0))
d1_err = d1_metric(dispPred, disp_true, (disp_true < args.maxdisp) & (disp_true > 0))
mean_d1 += d1_err
mean_epe += epe_err
if args.verbose:
if mid:
print('[INFO] Processing: %d/%d %s D1: %.3f EPE: %.3f Time: %.3f' % (
idx + 1, len(test_left), test_left[idx].split('/')[-2], d1_err * 100, epe_err, time_per_img))
else:
print('[INFO] Processing: %d/%d %s D1: %.3f EPE: %.3f Time: %.3f' % (
idx + 1, len(test_left), test_left[idx].split('/')[-1], d1_err * 100, epe_err, time_per_img))
if args.savepath is not None:
if kitti:
fileName = test_left[idx].split('/')[-1]
elif mid:
fileName = test_left[idx].split('/')[-2] + '_' + test_left[idx].split('/')[-1]
else:
if not osp.exists(osp.join(save_path, test_left[idx].split('/')[-4], test_left[idx].split('/')[-3])):
os.mkdir(osp.join(save_path, test_left[idx].split('/')[-4], test_left[idx].split('/')[-3]))
fileName = osp.join(test_left[idx].split('/')[-4], test_left[idx].split('/')[-3],
test_left[idx].split('/')[-1])
skimage.io.imsave(osp.join(save_path, fileName), (dispPred.numpy() * 256).astype('uint16'))
if mid:
print('[INFO] Mean D1: %.3f Mean EPE: %.3f' % (mean_d1 * 100 / 12.5, mean_epe / len(test_left)))
else:
print('[INFO] Mean D1: %.3f Mean EPE: %.3f' % (mean_d1 * 100 / len(test_left), mean_epe / len(test_left)))
print('[INFO] Average Time: %.3f' % (avg_time / len(test_left)))