-
-
Notifications
You must be signed in to change notification settings - Fork 73
/
test_cvusa.py
278 lines (243 loc) · 9.9 KB
/
test_cvusa.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
# -*- coding: utf-8 -*-
from __future__ import print_function, division
import argparse
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.autograd import Variable
import torch.backends.cudnn as cudnn
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import time
import os
import scipy.io
import yaml
import math
from model import ft_net, two_view_net, three_view_net
from utils import load_network
#fp16
try:
from apex.fp16_utils import *
except ImportError: # will be 3.x series
print('This is not an error. If you want to use low precision, i.e., fp16, please install the apex with cuda support (https://github.com/NVIDIA/apex) and update pytorch to 1.0')
######################################################################
# Options
# --------
parser = argparse.ArgumentParser(description='Training')
parser.add_argument('--gpu_ids',default='0', type=str,help='gpu_ids: e.g. 0 0,1,2 0,2')
parser.add_argument('--which_epoch',default='last', type=str, help='0,1,2,3...or last')
parser.add_argument('--test_dir',default='./data/cvpr2017_cvusa/val',type=str, help='./test_data')
parser.add_argument('--name', default='three_view_long_share_d0.75_256_s1_google', type=str, help='save model path')
parser.add_argument('--pool', default='avg', type=str, help='avg|max')
parser.add_argument('--batchsize', default=64, type=int, help='batchsize')
parser.add_argument('--h', default=256, type=int, help='height')
parser.add_argument('--w', default=256, type=int, help='width')
parser.add_argument('--views', default=2, type=int, help='views')
parser.add_argument('--use_dense', action='store_true', help='use densenet121' )
parser.add_argument('--PCB', action='store_true', help='use PCB' )
parser.add_argument('--multi', action='store_true', help='use multiple query' )
parser.add_argument('--fp16', action='store_true', help='use fp16.' )
parser.add_argument('--ms',default='1', type=str,help='multiple_scale: e.g. 1 1,1.1 1,1.1,1.2')
opt = parser.parse_args()
###load config###
# load the training config
config_path = os.path.join('./model',opt.name,'opts.yaml')
with open(config_path, 'r') as stream:
config = yaml.load(stream, Loader=yaml.FullLoader)
opt.fp16 = config['fp16']
opt.use_dense = config['use_dense']
opt.use_NAS = config['use_NAS']
opt.use_vgg16 = config['use_vgg16']
opt.stride = config['stride']
opt.views = config['views']
if 'h' in config:
opt.h = config['h']
opt.w = config['w']
if 'nclasses' in config: # tp compatible with old config files
opt.nclasses = config['nclasses']
else:
opt.nclasses = 729
str_ids = opt.gpu_ids.split(',')
#which_epoch = opt.which_epoch
name = opt.name
test_dir = opt.test_dir
gpu_ids = []
for str_id in str_ids:
id = int(str_id)
if id >=0:
gpu_ids.append(id)
print('We use the scale: %s'%opt.ms)
str_ms = opt.ms.split(',')
ms = []
for s in str_ms:
s_f = float(s)
ms.append(math.sqrt(s_f))
# set gpu ids
if len(gpu_ids)>0:
torch.cuda.set_device(gpu_ids[0])
cudnn.benchmark = True
######################################################################
# Load Data
# ---------
#
# We will use torchvision and torch.utils.data packages for loading the
# data.
#
data_transforms = transforms.Compose([
transforms.Resize((opt.h, opt.w), interpolation=3),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
if opt.PCB:
data_transforms = transforms.Compose([
transforms.Resize((384,192), interpolation=3),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
data_dir = test_dir
if opt.multi:
image_datasets = {x: datasets.ImageFolder( os.path.join(data_dir,x) ,data_transforms) for x in ['gallery','query','multi-query']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=opt.batchsize,
shuffle=False, num_workers=16) for x in ['gallery','query','multi-query']}
else:
image_datasets = {x: datasets.ImageFolder( os.path.join(data_dir,x) ,data_transforms) for x in ['satellite', 'street']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=opt.batchsize,
shuffle=False, num_workers=16) for x in ['satellite','street']}
use_gpu = torch.cuda.is_available()
######################################################################
# Extract feature
# ----------------------
#
# Extract feature from a trained model.
#
def fliplr(img):
'''flip horizontal'''
inv_idx = torch.arange(img.size(3)-1,-1,-1).long() # N x C x H x W
img_flip = img.index_select(3,inv_idx)
return img_flip
def which_view(name):
if 'satellite' in name:
return 1
elif 'street' in name:
return 2
elif 'drone' in name:
return 3
else:
print('unknown view')
return -1
def extract_feature(model,dataloaders, view_index = 1):
features = torch.FloatTensor()
count = 0
for data in dataloaders:
img, label = data
n, c, h, w = img.size()
count += n
print(count)
ff = torch.FloatTensor(n,512).zero_().cuda()
for i in range(2):
if(i==1):
img = fliplr(img)
input_img = Variable(img.cuda())
for scale in ms:
if scale != 1:
# bicubic is only available in pytorch>= 1.1
input_img = nn.functional.interpolate(input_img, scale_factor=scale, mode='bilinear', align_corners=False)
if opt.views ==2:
if view_index == 1:
outputs, _ = model(input_img, None)
elif view_index ==2:
_, outputs = model(None, input_img)
elif opt.views ==3:
if view_index == 1:
outputs, _, _ = model(input_img, None, None)
elif view_index ==2:
_, outputs, _ = model(None, input_img, None)
elif view_index ==3:
_, _, outputs = model(None, None, input_img)
ff += outputs
# norm feature
if opt.PCB:
# feature size (n,2048,6)
# 1. To treat every part equally, I calculate the norm for every 2048-dim part feature.
# 2. To keep the cosine score==1, sqrt(6) is added to norm the whole feature (2048*6).
fnorm = torch.norm(ff, p=2, dim=1, keepdim=True) * np.sqrt(6)
ff = ff.div(fnorm.expand_as(ff))
ff = ff.view(ff.size(0), -1)
else:
fnorm = torch.norm(ff, p=2, dim=1, keepdim=True)
ff = ff.div(fnorm.expand_as(ff))
features = torch.cat((features,ff.data.cpu()), 0)
return features
def get_id(img_path):
camera_id = []
labels = []
paths = []
for path, v in img_path:
folder_name = os.path.basename(os.path.dirname(path))
labels.append(int(folder_name))
paths.append(path)
return labels, paths
######################################################################
# Load Collected data Trained model
print('-------test-----------')
model, _, epoch = load_network(opt.name, opt)
model.classifier.classifier = nn.Sequential()
model = model.eval()
print(model)
if use_gpu:
model = model.cuda()
# Extract feature
since = time.time()
#gallery_name = 'gallery_street'
#query_name = 'query_satellite'
gallery_name = 'satellite'
query_name = 'street'
which_gallery = which_view(gallery_name)
which_query = which_view(query_name)
print('%d -> %d:'%(which_query, which_gallery))
gallery_path = image_datasets[gallery_name].imgs
f = open('gallery_name.txt','w')
for p in gallery_path:
f.write(p[0]+'\n')
query_path = image_datasets[query_name].imgs
f = open('query_name.txt','w')
for p in query_path:
f.write(p[0]+'\n')
gallery_label, gallery_path = get_id(gallery_path)
query_label, query_path = get_id(query_path)
if __name__ == "__main__":
with torch.no_grad():
query_feature = extract_feature(model,dataloaders[query_name], which_query)
gallery_feature = extract_feature(model,dataloaders[gallery_name], which_gallery)
# For street-view image, we use the avg feature as the final feature.
'''
if which_query == 2:
new_query_label = np.unique(query_label)
new_query_feature = torch.FloatTensor(len(new_query_label) ,512).zero_()
for i, query_index in enumerate(new_query_label):
new_query_feature[i,:] = torch.sum(query_feature[query_label == query_index, :], dim=0)
query_feature = new_query_feature
fnorm = torch.norm(query_feature, p=2, dim=1, keepdim=True)
query_feature = query_feature.div(fnorm.expand_as(query_feature))
query_label = new_query_label
elif which_gallery == 2:
new_gallery_label = np.unique(gallery_label)
new_gallery_feature = torch.FloatTensor(len(new_gallery_label), 512).zero_()
for i, gallery_index in enumerate(new_gallery_label):
new_gallery_feature[i,:] = torch.sum(gallery_feature[gallery_label == gallery_index, :], dim=0)
gallery_feature = new_gallery_feature
fnorm = torch.norm(gallery_feature, p=2, dim=1, keepdim=True)
gallery_feature = gallery_feature.div(fnorm.expand_as(gallery_feature))
gallery_label = new_gallery_label
'''
time_elapsed = time.time() - since
print('Test complete in {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
# Save to Matlab for check
result = {'gallery_f':gallery_feature.numpy(),'gallery_label':gallery_label,'gallery_path':gallery_path,'query_f':query_feature.numpy(),'query_label':query_label, 'query_path':query_path}
scipy.io.savemat('pytorch_result.mat',result)
print(opt.name)
result = './model/%s/result.txt'%opt.name
os.system('python evaluate_gpu.py | tee -a %s'%result)