-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtest_HSERGB.py
200 lines (160 loc) · 8.26 KB
/
test_HSERGB.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
import sys
sys.path.append('..')
from model import FusionModel
from tools import representation, event
from torchvision import transforms
import cv2
from PIL import Image
class HSERGBDataset:
def __init__(self, data_path='/home/lisiqi/data/HSERGB', subset='close', mode='train', folder='', number_of_frames_to_skip=15, nb_of_time_bin=20):
if mode not in ['train', 'test']:
raise ValueError
self.folder = os.path.join(data_path, subset, mode, folder)
self.number_of_frames_to_skip = number_of_frames_to_skip
print('skip %d frame' % self.number_of_frames_to_skip)
self.mode = mode
self.nb_of_time_bin = nb_of_time_bin
self.generate_data()
self.folder = folder
def __len__(self):
return len(self.idx)
def generate_data(self):
self.left_image = []
self.right_image = []
self.gt_image = []
self.event = []
self.gt_timestamp = []
self.idx = []
self.lr_timestamp = []
with open(os.path.join(self.folder, 'images_corrected', 'timestamp.txt'), 'r') as f:
ts = [float(l.strip('\n')) for l in f.readlines()]
N = len(ts)
for k in range(int(N/(self.number_of_frames_to_skip+1))-2):
rand = 0
start = k*(self.number_of_frames_to_skip+1) + rand
end = (k+1)*(self.number_of_frames_to_skip+1) + rand
self.left_image.append(os.path.join(self.folder, 'images_corrected', '%d.png'%start))
self.right_image.append(os.path.join(self.folder, 'images_corrected', '%d.png'%end))
self.event.append([os.path.join(self.folder, 'events_aligned', '%d.npz'%k) for k in range(start, end+1)])
self.gt_image.append([os.path.join(self.folder, 'images_corrected', '%d.png'%k) for k in range(start+1, end)])
self.gt_timestamp.append([ts[k] for k in range(start+1, end)])
self.lr_timestamp.append([ts[start], ts[end]])
for k in range(len(self.left_image)):
self.idx += [k] * len(self.gt_image[0])
self.start_idx = [k * len(self.gt_image[0]) for k in range(len(self.left_image))]
def __getitem__(self, idx):
seq_idx = self.idx[idx]
sample_idx = idx - self.start_idx[seq_idx]
left_image = transforms.ToTensor()(Image.open(self.left_image[seq_idx]))
right_image = transforms.ToTensor()(Image.open(self.right_image[seq_idx]))
w, h = left_image.shape[2], left_image.shape[1]
gt_image = transforms.ToTensor()(Image.open(self.gt_image[seq_idx][sample_idx]))
events = event.EventSequence.from_npz_files(self.event[seq_idx], h, w)
ts = self.gt_timestamp[seq_idx][sample_idx]
duration_left = ts - self.lr_timestamp[seq_idx][0]
duration_right = self.lr_timestamp[seq_idx][1] - ts
e_left = events.filter_by_timestamp(events.start_time(), duration_left)
e_right = events.filter_by_timestamp(ts, duration_right)
event_left_forward = representation.to_count_map(e_left, self.nb_of_time_bin).clone()
event_right_forward = representation.to_count_map(e_right, self.nb_of_time_bin).clone()
left_voxel_grid = representation.to_voxel_grid(e_left, nb_of_time_bins=5)
right_voxel_grid = representation.to_voxel_grid(e_right, nb_of_time_bins=5)
e_right.reverse()
e_left.reverse()
event_left_backward = representation.to_count_map(e_left, self.nb_of_time_bin)
event_right_backward = representation.to_count_map(e_right, self.nb_of_time_bin)
events_forward = np.concatenate((event_left_forward, event_right_forward), axis=-1)
events_backward = np.concatenate((event_right_backward, event_left_backward), axis=-1)
weight = duration_left / (duration_left+duration_right)
surface = events.filter_by_timestamp(ts-200, 400)
surface = representation.to_count_map(surface)
return events_forward, events_backward, left_image, right_image, gt_image, weight, \
[self.nb_of_time_bin, self.nb_of_time_bin], surface, left_voxel_grid, right_voxel_grid, self.gt_image[seq_idx][sample_idx]
def showMessage(message, file):
print(message)
with open(file, 'a') as f:
f.writelines(message + '\n')
def saveImg(img, path):
img[img > 1] = 1
img[img < 0] = 0
img = np.array(img[0].cpu() * 255)
img1 = np.zeros([img.shape[1], img.shape[2], 3])
img1[:,:,0] = img[2,:,:]
img1[:,:,1] = img[1,:,:]
img1[:,:,2] = img[0,:,:]
cv2.imwrite(path, img1)
if __name__ == '__main__':
import os
os.environ["CUDA_VISIBLE_DEVICES"] = '7'
import numpy as np
import torch
from torch import nn
from model import Metric
subset = 'close' # 'far'
number_of_frames_to_skip = 7 # 30
ckpt_path = f'./ckpt/ckpt_HSERGB_{subset}_x{number_of_frames_to_skip}.pth'
data_path = '/home/lisiqi/data/HSERGB'
assert subset in ['close', 'far']
assert number_of_frames_to_skip in [7, 30]
if subset == 'close':
testList = ['confetti', 'fountain_bellevue2', 'water_bomb_eth_01', 'water_bomb_floor_01', 'spinning_plate',
'baloon_popping', 'candle', 'fountain_schaffhauserplatz_02', 'spinning_umbrella']
else:
testList = ['lake_01', 'bridge_lake_03', 'bridge_lake_01', 'lake_03', 'sihl_03',
'kornhausbruecke_letten_random_04']
device = 'cuda'
input_img_channel = 3
nb_of_time_bin = 8
netParams = {'Ts': 1, 'tSample': nb_of_time_bin * 2}
theta = [3, 5, 10]
tauSr = [1, 2, 4]
tauRef = [1, 2, 4]
scaleRef = [1, 1, 1]
tauRho = [1, 1, 10]
scaleRho = [1, 1, 10]
run_dir = os.path.split(ckpt_path)[0]
print('rundir:', run_dir)
opF = os.path.join(run_dir, f'HSERGB_x{number_of_frames_to_skip}')
os.makedirs(opF, exist_ok=True)
with open(os.path.join(opF, 'ckpt.txt'), 'w') as f:
f.writelines(ckpt_path+'\n')
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.fastest = True
testFolder = [HSERGBDataset(data_path, subset, 'test', k, number_of_frames_to_skip, nb_of_time_bin) for k in testList]
testLoader = [torch.utils.data.DataLoader(testFolder[k], batch_size=1, shuffle=False, pin_memory=False, num_workers=1) for k in range(len(testFolder))]
total = sum([len(testLoader[k]) for k in range(len(testLoader))])
model = FusionModel(netParams, hidden_number=32, theta=theta, tauSr=tauSr, tauRef=tauRef, scaleRef=scaleRef,
tauRho=tauRho, scaleRho=scaleRho, channel=input_img_channel, fast_ckpt='', slow_ckpt='')
model = nn.DataParallel(model)
model = model.to(device)
print('==> loading existing model:', ckpt_path)
model_info = torch.load(ckpt_path)
model.load_state_dict(model_info['state_dict'])
l1 = torch.nn.L1Loss()
testMetirc = Metric()
with torch.no_grad():
model.eval()
count = 0
for loader in testLoader:
opFolder = os.path.join(opF, subset, loader.dataset.folder, 'test')
os.makedirs(opFolder, exist_ok=True)
for i, (events_forward, events_backward, left_image, right_image, gt_image, weight, [n_left, n_right],
surface, left_voxel_grid, right_voxel_grid, name) in enumerate(loader):
events_forward = events_forward.cuda()
events_backward = events_backward.cuda()
left_image = left_image.cuda()
right_image = right_image.cuda()
gt_image = gt_image.cuda()
weight = weight.cuda()
surface = surface.cuda()
left_voxel_grid = left_voxel_grid.cuda()
right_voxel_grid = right_voxel_grid.cuda()
output = model(events_forward, events_backward, left_image, right_image, weight, n_left, n_right, surface, left_voxel_grid, right_voxel_grid)
L1Loss = l1(output, gt_image)
testMetirc.update(L1Loss=L1Loss, Lpips=torch.tensor([0]), FeaLoss=torch.tensor([0]), total=L1Loss)
saveImg(output, os.path.join(opFolder, '%06d.png'%int(name[0].split('/')[-1].strip('.png'))))
if count % 50 == 0:
avg = testMetirc.get_average_epoch()
message = 'Test, Iter [%d]/[%d], l1:%f' % (count, total, avg[0])
print(message)
count += 1