-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
316 lines (247 loc) · 12.7 KB
/
train.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
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
import argparse
import os
import numpy as np
import math
import itertools
from torch.utils.tensorboard import SummaryWriter
import torchvision.transforms as transforms
# from torch.optim.lr_scheduler import LambdaLR
from torch.optim.lr_scheduler import ExponentialLR
from torch.utils.data import DataLoader
from tqdm import tqdm
import torch
import torch.nn as nn
import torch.nn.functional as F
from model import SatelliteFeatureExtractor, StreetFeatureExtractor, TransMixer
from dataset import ImageDataset
from SMTL import softMarginTripletLoss
STREET_IMG_WIDTH = 320
STREET_IMG_HEIGHT = 180
SATELLITE_IMG_WIDTH = 256
SATELLITE_IMG_HEIGHT = 256
SEQUENCE_SIZE = 7
def ValidateOne(distArray, topK):
acc = 0.0
dataAmount = 0.0
for i in range(distArray.shape[0]):
groundTruths = distArray[i,i]
pred = torch.sum(distArray[:,i] < groundTruths)
if pred < topK:
acc += 1.0
dataAmount += 1.0
return acc / dataAmount
def ValidateAll(streetFeatures, satelliteFeatures):
distArray = 2 - 2 * torch.matmul(satelliteFeatures, torch.transpose(streetFeatures, 0, 1))
topOnePercent = int(distArray.shape[0] * 0.01) + 1
valAcc = torch.zeros((1, topOnePercent))
for i in range(topOnePercent):
valAcc[0,i] = ValidateOne(distArray, i)
return valAcc
class LambdaLR():
def __init__(self, n_epochs, offset, decay_start_epoch):
'''
linear decay LR scheduler
n_epochs: number of total training epochs
offset: train start epochs
decay_start_epoch: epoch start decay
'''
assert ((n_epochs - decay_start_epoch) > 0), "Decay must start before the training session ends!"
self.n_epochs = n_epochs
self.offset = offset
self.decay_start_epoch = decay_start_epoch
def step(self, epoch):
return 1.0 - max(0, epoch + self.offset - self.decay_start_epoch)/(self.n_epochs - self.decay_start_epoch)
def save_model(savePath, transMixer, sateFeature, strFeature, epoch):
modelFolder = os.path.join(savePath, f"epoch_{epoch}")
os.makedirs(modelFolder)
torch.save(transMixer.state_dict(), os.path.join(modelFolder, f'trans_{epoch}.pth'))
torch.save(sateFeature.state_dict(), os.path.join(modelFolder, f'SFE_{epoch}.pth'))
torch.save(strFeature.state_dict(), os.path.join(modelFolder, f'GFE_{epoch}.pth'))
# torch.save(HPEstimator.state_dict(), os.path.join(modelFolder, f'HPE_{epoch}.pth'))
def InferOnce(grdFE, satFE, transMixer, batch, device, noMask):
grdImgs = batch["street"].to(device)
sateImgs = batch["satellite"].to(device)
numSeqInBatch = grdImgs.shape[0]
#street view featuer extraction
grdImgs = grdImgs.view(grdImgs.shape[0]*grdImgs.shape[1],\
grdImgs.shape[2],grdImgs.shape[3], grdImgs.shape[4])
grdFeature = grdFE(grdImgs)
grdFeature = grdFeature.view(numSeqInBatch, SEQUENCE_SIZE, -1)
#satellite view feature extraction
sateImgs = sateImgs.view(sateImgs.shape[0], sateImgs.shape[1]*sateImgs.shape[2],\
sateImgs.shape[3], sateImgs.shape[4])
sateFeature = satFE(sateImgs)
sateFeature = sateFeature.view(numSeqInBatch, -1)
# print(sateFeature.shape)
if not noMask:
grdMixedFeature = transMixer(grdFeature, mask=True, masked_range = [0,6], max_masked=opt.max_masked)
else:
grdMixedFeature = transMixer(grdFeature, mask=False, masked_range = [0,6])
grdGlobalFeature = grdMixedFeature.permute(0,2,1)
grdGlobalLatent = F.avg_pool1d(grdGlobalFeature, grdGlobalFeature.shape[2]).squeeze(2)
return sateFeature, grdGlobalLatent
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--n_epochs", type=int, default=50, help="number of epochs of training")
parser.add_argument("--epoch", type=int, default=0, help="number of epochs start training")
parser.add_argument("--decay_epoch", type=int, default=30, help="number of epochs start decaying LR")
parser.add_argument("--batch_size", type=int, default=24, help="size of the batches")
parser.add_argument("--lr", type=float, default=1e-5, help="learning rate")
parser.add_argument("--save_name", type=str, default='SAVE_NAME', help='name of the model')
parser.add_argument("--feature_dims", type=int, default=4096, help="latent feature dimension")
parser.add_argument("--backbone", type=str, default="vgg16", help='weight for heading loss')
parser.add_argument("--beta1", type=float, default=0.9, help='beta1 for adam')
parser.add_argument("--beta2", type=float, default=0.999, help='beta2 for adam')
parser.add_argument("--num_workers", type=int, default=12, help='num of CPUs')
parser.add_argument("--lambda_SMTL", type=float, default=1.0, help='weight for triplet loss')
parser.add_argument("--gamma", type=float, default=10.0, help='value for SMTL gamma')
parser.add_argument("--weight_decay", type=float, default=1e-2, help='value for SMTL gamma')
parser.add_argument('--no_mask', default=False, action='store_true')
parser.add_argument('--MHA_layers', type=int, default=6, help="number of MHA layers")
parser.add_argument('--max_masked', type=int, default=6, help="max masked frames")
parser.add_argument('--nHeads', type=int, default=8, help="number of heads")
opt = parser.parse_args()
print(opt)
zoom = 20
print(f"zoom level:{zoom}")
#saving path for training logs
writer = SummaryWriter(opt.save_name)
savePath=os.path.join(opt.save_name, 'training_logs')
if not os.path.exists(savePath):
os.makedirs(savePath)
else:
print("Note! Saving path existed !")
#set device
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("Using device:",device)
length = 7
print("sequence length : ", length)
transMixer = TransMixer(transDimension=opt.feature_dims, max_length=length, numLayers = opt.MHA_layers, nHead=opt.nHeads)
grdFeatureExtractor = StreetFeatureExtractor(backbone = opt.backbone)
satelliteFeatureExtractor = SatelliteFeatureExtractor(backbone = opt.backbone, inputChannel=3)
if torch.cuda.device_count() > 1:
transMixer = nn.DataParallel(transMixer)
grdFeatureExtractor = nn.DataParallel(grdFeatureExtractor)
satelliteFeatureExtractor = nn.DataParallel(satelliteFeatureExtractor)
#all networks to cuda if available
transMixer.to(device)
grdFeatureExtractor.to(device)
satelliteFeatureExtractor.to(device)
# Optimizers
optimizer = torch.optim.Adam(itertools.chain(transMixer.parameters(),grdFeatureExtractor.parameters(), satelliteFeatureExtractor.parameters()), lr=opt.lr, betas=(opt.beta1, opt.beta2), weight_decay=1e-6)
lrSchedule = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=LambdaLR(opt.n_epochs, opt.epoch, opt.decay_epoch).step)
#data loader
transformsSatellite = [transforms.Resize((SATELLITE_IMG_HEIGHT, SATELLITE_IMG_WIDTH)),
transforms.ColorJitter(0.1, 0.1, 0.1),
transforms.ToTensor(),
transforms.RandomErasing(p=0.1, scale=(0.1,0.2),value="random"),
transforms.Normalize([0.5,0.5,0.5], [0.5,0.5,0.5])
]
transformsStreet = [transforms.Resize((STREET_IMG_HEIGHT, STREET_IMG_WIDTH)),
transforms.ColorJitter(0.1, 0.1, 0.1),
transforms.ToTensor(),
transforms.RandomErasing(p=0.1, scale=(0.1,0.2),value="random"),
transforms.Normalize([0.5,0.5,0.5], [0.5,0.5,0.5])
]
transformsStreetVal = [transforms.Resize((STREET_IMG_HEIGHT, STREET_IMG_WIDTH)),
transforms.ToTensor(),
transforms.Normalize([0.5,0.5,0.5], [0.5,0.5,0.5])
]
transformsSatelliteVal = [transforms.Resize((SATELLITE_IMG_HEIGHT, SATELLITE_IMG_WIDTH)),
transforms.ToTensor(),
transforms.Normalize([0.5,0.5,0.5], [0.5,0.5,0.5])
]
trainLoader = DataLoader(ImageDataset(transforms_street=transformsStreet,transforms_sat=transformsSatellite,mode="train",zoom=zoom),\
batch_size=opt.batch_size, shuffle=True, num_workers= opt.num_workers)
valLoader = DataLoader(ImageDataset(transforms_street=transformsStreetVal,transforms_sat=transformsSatelliteVal,mode="val",zoom=zoom),\
batch_size=opt.batch_size, shuffle=False, num_workers= opt.num_workers)
##training
allLosses = []
print("start training...")
for epoch in range(opt.n_epochs):
#set the model to train mode
transMixer.train()
grdFeatureExtractor.train()
satelliteFeatureExtractor.train()
epochLoss = 0
epochTripletLoss = 0
for batch in tqdm(trainLoader, disable = False):
if batch["street"].shape[0] < 2:
continue
sateFeature, grdGlobalLatent =\
InferOnce(grdFeatureExtractor,\
satelliteFeatureExtractor, \
transMixer, batch, device, opt.no_mask)
#softmargin triplet loss
sateFeatureUnit = sateFeature / torch.linalg.norm(sateFeature,dim=1,keepdim=True)
grdGlobalLatentUnit = grdGlobalLatent / torch.linalg.norm(grdGlobalLatent,dim=1,keepdim=True)
lossTriplet = softMarginTripletLoss(sateFeatureUnit, grdGlobalLatentUnit, opt.gamma)
loss = opt.lambda_SMTL * lossTriplet
#optimize
optimizer.zero_grad()
loss.backward()
# torch.nn.utils.clip_grad_norm_(itertools.chain(transMixer.parameters(),grdFeatureExtractor.parameters(), satelliteFeatureExtractor.parameters()), 1.0)
optimizer.step()
epochLoss += loss.item()
epochTripletLoss += lossTriplet.item()
#step learning rate
lrSchedule.step()
#calculate epoch average loss
epochLoss = float(epochLoss) / float(len(trainLoader))
epochTripletLoss = float(epochTripletLoss) / float(len(trainLoader))
#add to all losses list
allLosses.append(epochLoss)
if epoch % 10 == 9:
save_model(savePath, transMixer, satelliteFeatureExtractor,\
grdFeatureExtractor, epoch+1)
#set the model to evaluate mode
transMixer.eval()
grdFeatureExtractor.eval()
satelliteFeatureExtractor.eval()
valSateFeatures = None
valStreetFeature = None
with torch.no_grad():
for batch in tqdm(valLoader, disable = False):
sateFeature, grdGlobalLatent=\
InferOnce(grdFeatureExtractor,\
satelliteFeatureExtractor, \
transMixer, batch, device, opt.no_mask)
#softmargin triplet loss
sateFeatureUnit = sateFeature / torch.linalg.norm(sateFeature,dim=1,keepdim=True)
grdGlobalRespresentUnit = grdGlobalLatent / torch.linalg.norm(grdGlobalLatent,dim=1,keepdim=True)
#stack features to the container
if valSateFeatures == None:
valSateFeatures = sateFeatureUnit.detach()
else:
valSateFeatures = torch.cat((valSateFeatures, sateFeatureUnit.detach()), dim=0)
if valStreetFeature == None:
valStreetFeature = grdGlobalRespresentUnit.detach()
else:
valStreetFeature = torch.cat((valStreetFeature, grdGlobalRespresentUnit.detach()), dim=0)
#Retrival accuracy
valAcc = ValidateAll(valStreetFeature, valSateFeatures)
print(f"==============Summary of epoch {epoch} on validation set=================")
try:
#print epoch loss
print("---------loss---------")
print(f"Epoch {epoch} Loss {epochLoss}")
print(f"triplet loss:{epochTripletLoss}")
writer.add_scalars('losses',{
'epoch_loss':epochLoss,
'triplet_loss':epochTripletLoss
}, epoch)
print("----------------------")
print('top1', ':', valAcc[0, 1] * 100.0)
print('top5', ':', valAcc[0, 5] * 100.0)
print('top10', ':', valAcc[0, 10] * 100.0)
print('top1%', ':', valAcc[0, -1] * 100.0)
writer.add_scalars('validation recall@k',{
'top 1':valAcc[0, 1],
'top 5':valAcc[0, 5],
'top 10':valAcc[0, 10],
'top 1%':valAcc[0, -1]
}, epoch)
except:
print(valAcc)
print("========================================================================")
writer.close()#close tensorboard