-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_patchnet.py
332 lines (292 loc) · 15.9 KB
/
train_patchnet.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
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
import torch
import argparse, os
import numpy as np
import random
import math
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms
from lr_scheduler import PolyScheduler
import albumentations as A
from albumentations.pytorch import ToTensorV2 as ToTensor
from torchvision import models
from torchvision.models import ResNet18_Weights
from models.CDCNs import Conv2d_cd, CDCNpp
from models.AdMSLoss import AdMSoftmaxLoss
from datasets.oulup_dataset import AMOuluDataset
from datasets.lcc_fasd import LccFasdDataset
import torch.nn.functional as F
import torch.optim as optim
import copy
import torch.nn as nn
from utils import performances
def setup_seed(seed, cuda_deterministic=True):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
os.environ["PYTHONHASHSEED"] = str(seed)
if cuda_deterministic: # slower, more reproducible
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
else: # faster, less reproducible
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.benchmark = True
lcc_transform = transforms.Compose([
transforms.ToPILImage(),
transforms.Resize((160, 160)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]),])
def transform(mode):
if mode == 'train':
transform = A.Compose(
[
A.RandomCrop(height=160, width=160, p=1.0),
A.HorizontalFlip(p=0.5),
# A.VerticalFlip(p=0.5),
A.RandomRotate90(p=0.5),
# A.augmentations.transforms.ISONoise(color_shift=(0.15,0.35),
# intensity=(0.2, 0.5), p=0.5), #p=0.2),
# A.augmentations.transforms.RandomBrightnessContrast(brightness_limit=0.2,
# contrast_limit=0.2,
# brightness_by_max=True,
# always_apply=False, p=0.5), #p=0.3),
# A.MotionBlur(blur_limit=5, p=0.5), #p=0.2),
# A.RGBShift(r_shift_limit=15, g_shift_limit=15, b_shift_limit=15, p=0.5),
# A.ShiftScaleRotate(shift_limit=0.05, scale_limit=0.0, rotate_limit=45, p=0.5),
A.transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)),
ToTensor(),
]
)
else:
transform = A.Compose(
[
A.Resize(height=160, width=160, p=1.0),
A.transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)),
ToTensor(),
]
)
return transform
def train_test():
writer = SummaryWriter(log_dir='./runs')
setup_seed(2048, cuda_deterministic=False)
isExists = os.path.exists(args.log)
if not isExists:
os.makedirs(args.log)
log_file = open(args.log+'/'+ args.log+'_log_P1.txt', 'w')
print("Oulu-NPU, P1:\n")
log_file.write('Oulu-NPU, PatchNet:\n')
log_file.flush()
# GPU & log file --> if use DataParallel, please comment this command
# os.environ["CUDA_VISIBLE_DEVICES"] = "%d" % (args.gpu)
device = [0, 1]
# dataset
train_data = AMOuluDataset(root_folder='/mnt/training_dataset/face_dataset/Oulu_align',\
mode='train',\
transform=transform(mode='train'),)
train_loader = DataLoader(train_data, batch_size=args.batchsize, shuffle=True, num_workers=4, pin_memory=True, )
valid_data = AMOuluDataset(root_folder='/mnt/training_dataset/face_dataset/Oulu_align',\
mode='valid',\
transform=transform(mode='valid'),)
valid_loader = DataLoader(valid_data, batch_size=64, shuffle=True, num_workers=4)
# model
# model = CDCNpp(basic_conv=Conv2d_cd, theta=0.7, num_class=30)
model = models.resnet18(weights=ResNet18_Weights.DEFAULT)
model.fc = nn.Identity()
AM_model = AdMSoftmaxLoss(in_features=512, out_features=30)
model = model.cuda()
# model.load_state_dict(torch.load('models/model.pt'))
# model = model.to(device[0])
# model = nn.DataParallel(model, device_ids=device, output_device=device[0])
AM_model = AM_model.cuda()
# AM_model.load_state_dict(torch.load('models/AM_model.pt'))
# AM_model = AM_model.to(device[0])
# AM_model = nn.DataParallel(AM_model, device_ids=device, output_device=device[0])
# print('load weight !!')
params = list(model.parameters()) + list(AM_model.parameters())
optimizer = optim.SGD(params, lr=args.lr, momentum=0.9, weight_decay=0.0005)
# # scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=args.step_size, gamma=args.gamma)
print(f"train loader size: {len(train_loader)}")
total_step = len(train_loader) * (args.epochs)
lr_scheduler = PolyScheduler(
optimizer=optimizer,
base_lr=args.lr,
max_steps=total_step,
warmup_steps=0,
last_epoch=-1
)
# #bandpass_filter_numpy = build_bandpass_filter_numpy(30, 30) # fs, order # 61, 64
model_save_path = './models/model.pt'
AM_model_save_path = './models/AM_model.pt'
best_ACER = math.inf
best_model_wts = copy.deepcopy(model.state_dict())
best_AMmodel_wts = copy.deepcopy(AM_model.state_dict())
train_idx = 0
val_idx = 0
for epoch in range(args.epochs): # loop over the dataset multiple times
# loss_absolute = AvgrageMeter()
# loss_contra = AvgrageMeter()
###########################################
''' train '''
###########################################
model.train()
AM_model.train()
for i, (data1, data2, labels, binary) in enumerate(train_loader):
# with torch.autograd.detect_anomaly():
# get the inputs
inputs1, inputs2, labels, binary = data1.cuda(), data2.cuda(), labels.cuda(), binary.cuda()
optimizer.zero_grad()
# forward + backward + optimize
# feat_vector1, map_x1 = model(inputs1)
# feat_vector2, map_x2 = model(inputs2)
feat_vector1 = model(inputs1)
feat_vector2 = model(inputs2)
# self-supervised similarity
feat_vector1 = F.normalize(feat_vector1, dim=1)
feat_vector2 = F.normalize(feat_vector2, dim=1)
simi_loss = torch.mean(torch.norm(feat_vector1-feat_vector2, dim=1))
# AM-Softmax loss
AM_loss1 = AM_model(feat_vector1, labels)
AM_loss2 = AM_model(feat_vector2, labels)
AM_loss = AM_loss1 + AM_loss2
# Full loss
loss = simi_loss + AM_loss
# print(loss)
# eps = 1e-6
# if loss.isnan():
# print(inputs1)
# print(inputs2)
# break
# continue
# else: loss = loss
# Accuracy
# live_prob1 = AM_model._predict(feat_vector1)
# live_prob2 = AM_model._predict(feat_vector2)
# y_1 = torch.ones(16)
# y_2 = torch.ones(16)
# real_1 = live_prob1 >= 0.5
# real_2 = live_prob2 >= 0.5
# real1_indices = real_1.nonzero().squeeze(1)
# real2_indices = real_2.nonzero().squeeze(1)
# y_1[real1_indices] = 0
# y_2[real2_indices] = 0
# acc1 = (y_1.cuda() == binary).float().mean()
# acc2 = (y_2.cuda() == binary).float().mean()
# acc = (acc1+acc2)/2
writer.add_scalar('Loss/Train', loss, train_idx)
# writer.add_scalar('Acc/Train', acc, train_idx)
train_idx += 1
loss.backward()
optimizer.step()
lr_scheduler.step()
if i % args.echo_batches == 0: # print every 50 mini-batches
# log written
print('epoch:%d, mini-batch:%3d, lr=%.7f, simi_loss= %.4f, AM_loss= %.4f'% (epoch + 1, i, float(lr_scheduler.get_last_lr()[0]), simi_loss, AM_loss))
log_file.write('epoch:%d, mini-batch:%3d, lr=%.7f, simi_loss= %.4f, AM_loss= %.4f \n' % (epoch + 1, i, float(lr_scheduler.get_last_lr()[0]), simi_loss, AM_loss))
log_file.flush()
for name, param in AM_model.named_parameters():
# writer.add_histogram(tag=name+'_grad', values=param.grad, global_step=epoch)
writer.add_histogram(tag=name+'_data', values=param.data, global_step=epoch+1)
# whole epoch average
# print('epoch:%d, Train: Absolute_Depth_loss= %.4f, Contrastive_Depth_loss= %.4f\n' % (epoch + 1, loss_absolute.avg, loss_contra.avg))
# log_file.write('epoch:%d, Train: Absolute_Depth_loss= %.4f, Contrastive_Depth_loss= %.4f \n' % (epoch + 1, loss_absolute.avg, loss_contra.avg))
# log_file.flush()
if (epoch+1) % 5 == 0: # test every 5 epochs
model.eval()
AM_model.eval()
with torch.no_grad():
###########################################
''' val '''
###########################################
# val for threshold
score_list = []
print('------Start Validation------')
log_file.write('------Start Validation------\n')
log_file.flush()
for i, (inputs, labels, binary) in enumerate(valid_loader):
# get the inputs
inputs, labels, binary = inputs, labels.cuda(), binary.cuda()
optimizer.zero_grad()
if i % 500:
print(f'index: {i}')
# forward + backward + optimize
for j in range(len(inputs)):
img_i = inputs[j].cuda()
# feat_vectori, map_xi = model(img_i)
feat_vectori = model(img_i)
live_prob = AM_model._predict(feat_vectori)
if j == 0:
sum_prob = live_prob
else:
sum_prob += live_prob
avg_prob = sum_prob / 9
for k in range(int(avg_prob.shape[0])):
live_prob = avg_prob[k]
score_list.append('{} {}\n'.format(live_prob, int(binary[k])))
map_score_val_filename = args.log+'/'+ args.log+'_map_score_val.txt'
with open(map_score_val_filename, 'w') as file:
file.writelines(score_list)
# ###########################################
# ''' test '''
# ##########################################
# # test for ACC
# test_data = LccFasdDataset(root_dir='/mnt/training_dataset/face_dataset/LCC_FASD',\
# protocol='combine_all',\
# transform=lcc_transform,\
# get_img_path=False)
# dataloader_test = DataLoader(test_data, batch_size=1, shuffle=True, num_workers=4)
# map_score_list = []
# print('------Start Test------')
# for i, (inputs, labels) in enumerate(dataloader_test):
# # get the inputs
# inputs, spoof_label = inputs.cuda(), labels.cuda()
# map_score = 0.0
# for frame_t in range(inputs.shape[0]):
# map_x, embedding, x_Block1, x_Block2, x_Block3, x_input = model(inputs)
# score_norm = torch.sum(map_x)/1024
# map_score += score_norm
# map_score = map_score/inputs.shape[0]
# map_score_list.append('{} {}\n'.format(map_score, int(spoof_label)))
# map_score_test_filename = args.log+'/'+ args.log+'_map_score_test.txt'
# with open(map_score_test_filename, 'w') as file:
# file.writelines(map_score_list)
# #############################################################
# # performance measurement both val and test
# #############################################################
val_threshold, val_ACC, val_AUC, val_APCER, val_BPCER, val_ACER = performances(map_score_val_filename)
print('epoch:%d, Val: val_threshold= %.4f, val_ACC= %.4f, val_AUC= %.4f, val_APCER= %.4f, val_BPCER= %.4f, val_ACER= %.4f' % (epoch + 1, val_threshold, val_ACC, val_AUC, val_APCER, val_BPCER, val_ACER))
log_file.write('epoch:%d, Val: val_threshold= %.4f, val_ACC= %.4f, val_AUC= %.4f, val_APCER= %.4f, val_BPCER= %.4f, val_ACER= %.4f \n' % (epoch + 1, val_threshold, val_ACC, val_AUC, val_APCER, val_BPCER, val_ACER))
log_file.flush()
# print('epoch:%d, Test: ACC= %.4f, APCER= %.4f, BPCER= %.4f, ACER= %.4f' % (epoch + 1, test_ACC, test_APCER, test_BPCER, test_ACER))
# #print('epoch:%d, Test: test_threshold= %.4f, test_ACER_test_threshold= %.4f\n' % (epoch + 1, test_threshold, test_ACER_test_threshold))
if val_ACER < best_ACER:
best_ACER = val_ACER
best_model_wts = copy.deepcopy(model.state_dict())
best_AMmodel_wts = copy.deepcopy(AM_model.state_dict())
torch.save(model.state_dict(), model_save_path)
torch.save(AM_model.state_dict(), AM_model_save_path)
print('...Saving model with ACER: {:.4f}'.format(val_ACER))
log_file.write('...Saving model with ACER: {:.4f} \n'.format(val_ACER))
log_file.flush()
# log_file.write('epoch:%d, Test: ACC= %.4f, APCER= %.4f, BPCER= %.4f, ACER= %.4f \n' % (epoch + 1, test_ACC, test_APCER, test_BPCER, test_ACER))
# #log_file.write('epoch:%d, Test: test_threshold= %.4f, test_ACER_test_threshold= %.4f \n\n' % (epoch + 1, test_threshold, test_ACER_test_threshold))
# log_file.flush()
writer.add_scalar('Metrics/vACER', val_ACER, val_idx)
# writer.add_scalar('Metrics/tACER', test_ACER, val_idx)
# writer.add_scalar('Metrics/tAPCER', test_APCER, val_idx)
# writer.add_scalar('Metrics/tBPCER', test_BPCER, val_idx)
print('Finished Training')
if __name__ == "__main__":
torch.backends.cudnn.benchmark = True
parser = argparse.ArgumentParser(description="save quality using landmarkpose model")
parser.add_argument('--gpu', type=int, default=3, help='the gpu id used for predict')
parser.add_argument('--lr', type=float, default=0.01, help='initial learning rate')
parser.add_argument('--batchsize', type=int, default=64, help='initial batchsize')
# parser.add_argument('--step_size', type=int, default=100, help='how many epochs lr decays once') # 500
# parser.add_argument('--gamma', type=float, default=0.5, help='gamma of optim.lr_scheduler.StepLR, decay of lr')
parser.add_argument('--echo_batches', type=int, default=50, help='how many batches display once') # 50
parser.add_argument('--epochs', type=int, default=200, help='total training epochs')
parser.add_argument('--log', type=str, default="CDCNpp_patch", help='log and save model name')
parser.add_argument('--finetune', action='store_true', default=False, help='whether finetune other models')
args = parser.parse_args()
train_test()