forked from xuexingyu24/MobileFaceNet_Tutorial_Pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathTrain.py
182 lines (148 loc) · 8.56 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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu May 23 08:57:15 2019
Train Mobilefacenet
@author: AIRocker
"""
import os
import sys
sys.path.append('..')
import numpy as np
import argparse
import torch
import torchvision.transforms as transforms
import torch.utils.data as data
import torch.optim as optim
from torch.optim import lr_scheduler
from data_set.dataloader import LFW, CFP_FP, AgeDB30, CASIAWebFace, MS1M
from face_model import MobileFaceNet, Arcface
import time
from Evaluation import getFeature, evaluation_10_fold
def load_data(batch_size, dataset = 'Faces_emore'):
transform = transforms.Compose([
transforms.ToTensor(), # range [0, 255] -> [0.0,1.0]
transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))]) # range [0.0, 1.0] -> [-1.0,1.0]
root = 'data_set/LFW/lfw_align_112'
file_list = 'data_set/LFW/pairs.txt'
dataset_LFW = LFW(root, file_list, transform=transform)
root = 'data_set/CFP-FP/CFP_FP_aligned_112'
file_list = 'data_set/CFP-FP/cfp_fp_pair.txt'
dataset_CFP_FP = CFP_FP(root, file_list, transform=transform)
root = 'data_set/AgeDB-30/agedb30_align_112'
file_list = 'data_set/AgeDB-30/agedb_30_pair.txt'
dataset_AgeDB30 = AgeDB30(root, file_list, transform=transform)
if dataset == 'CASIA':
root = 'data_set/CASIA_Webface_Image'
file_list = 'data_set/CASIA_Webface_Image/webface_align_112.txt'
dataset_train = CASIAWebFace(root, file_list, transform=transform)
elif dataset == 'Faces_emore':
root = 'data_set/faces_emore_images'
file_list = 'data_set/faces_emore_images/faces_emore_align_112.txt'
dataset_train = MS1M(root, file_list, transform=transform)
else:
raise NameError('no training data exist!')
dataloaders = {'train': data.DataLoader(dataset_train, batch_size=batch_size, shuffle=True, num_workers=2),
'LFW': data.DataLoader(dataset_LFW, batch_size=batch_size, shuffle=False, num_workers=2),
'CFP_FP': data.DataLoader(dataset_CFP_FP, batch_size=batch_size, shuffle=False, num_workers=2),
'AgeDB30': data.DataLoader(dataset_AgeDB30, batch_size=batch_size, shuffle=False, num_workers=2)}
dataset = {'train': dataset_train,'LFW': dataset_LFW,
'CFP_FP': dataset_CFP_FP, 'AgeDB30': dataset_AgeDB30}
dataset_sizes = {'train': len(dataset_train), 'LFW': len(dataset_LFW),
'CFP_FP': len(dataset_CFP_FP), 'AgeDB30': len(dataset_AgeDB30)}
print('training and validation data loaded')
return dataloaders, dataset_sizes, dataset
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Face_Detection_Training')
parser.add_argument('--dataset', type=str, default='Faces_emore', help='Training dataset: CASIA, Faces_emore')
parser.add_argument('--feature_dim', type=int, default=512, help='the feature dimension output')
parser.add_argument('--batch_size', type=int, default=128, help='batch size for training and evaluation')
parser.add_argument('--epoch', type=int, default=12, help='number of epoches for training')
parser.add_argument('--method', type=str, default='l2_distance',
help='methold to evaluate feature similarity, l2_distance, cos_distance')
parser.add_argument('--flip', type=str, default=True, help='if flip the image with time augmentation')
args = parser.parse_args()
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
dataloaders , dataset_sizes, dataset = load_data(args.batch_size, dataset = args.dataset)
model = MobileFaceNet(args.feature_dim).to(device) # embeding size is 512 (feature vector)
print('MobileFaceNet face detection model loaded')
margin = Arcface(embedding_size=args.feature_dim, classnum=int(dataset['train'].class_nums), s=32., m=0.5).to(device)
criterion = torch.nn.CrossEntropyLoss().to(device)
optimizer_ft = optim.SGD([
{'params': model.parameters(), 'weight_decay': 5e-4},
{'params': margin.parameters(), 'weight_decay': 5e-4}], lr=0.01, momentum=0.9, nesterov=True)
exp_lr_scheduler = lr_scheduler.MultiStepLR(optimizer_ft, milestones=[6, 8, 10], gamma=0.3)
start = time.time()
## save logging and weights
train_logging_file = 'train_{}_logging.txt'.format(args.dataset)
test_logging_file = 'test_{}_logging.txt'.format(args.dataset)
save_dir = 'saving_{}_ckpt'.format(args.dataset)
if os.path.exists(save_dir):
raise NameError('model dir exists!')
os.makedirs(save_dir)
best_acc = {'LFW': 0.0, 'CFP_FP': 0.0, 'AgeDB30': 0.0}
best_iters = {'LFW': 0, 'CFP_FP': 0, 'AgeDB30': 0}
total_iters = 0
print('training kicked off..')
print('-' * 10)
for epoch in range(args.epoch):
# train model
exp_lr_scheduler.step()
model.train()
since = time.time()
for det in dataloaders['train']:
img, label = det[0].to(device), det[1].to(device)
optimizer_ft.zero_grad()
with torch.set_grad_enabled(True):
raw_logits = model(img)
output = margin(raw_logits, label)
loss = criterion(output, label)
loss.backward()
optimizer_ft.step()
total_iters += 1
# print train information
if total_iters % 100 == 0:
# current training accuracy
_, preds = torch.max(output.data, 1)
total = label.size(0)
correct = (np.array(preds.cpu()) == np.array(label.data.cpu())).sum()
time_cur = (time.time() - since) / 100
since = time.time()
for p in optimizer_ft.param_groups:
lr = p['lr']
print("Epoch {}/{}, Iters: {:0>6d}, loss: {:.4f}, train_accuracy: {:.4f}, time: {:.2f} s/iter, learning rate: {}"
.format(epoch, args.epoch-1, total_iters, loss.item(), correct/total, time_cur, lr))
with open(train_logging_file, 'a') as f:
f.write("Epoch {}/{}, Iters: {:0>6d}, loss: {:.4f}, train_accuracy: {:.4f}, time: {:.2f} s/iter, learning rate: {}"
.format(epoch, args.epoch-1, total_iters, loss.item(), correct/total, time_cur, lr)+'\n')
f.close()
# save model
if total_iters % 3000 == 0:
torch.save({
'iters': total_iters,
'net_state_dict': model.state_dict()},
os.path.join(save_dir, 'Iter_%06d_model.ckpt' % total_iters))
torch.save({
'iters': total_iters,
'net_state_dict': margin.state_dict()},
os.path.join(save_dir, 'Iter_%06d_margin.ckpt' % total_iters))
# evaluate accuracy
if total_iters % 3000 == 0:
model.eval()
for phase in ['LFW', 'CFP_FP', 'AgeDB30']:
featureLs, featureRs = getFeature(model, dataloaders[phase], device, flip = args.flip)
ACCs, threshold = evaluation_10_fold(featureLs, featureRs, dataset[phase], method = args.method)
print('Epoch {}/{}，{} average acc:{:.4f} average threshold:{:.4f}'
.format(epoch, args.epoch-1, phase, np.mean(ACCs) * 100, np.mean(threshold)))
if best_acc[phase] <= np.mean(ACCs) * 100:
best_acc[phase] = np.mean(ACCs) * 100
best_iters[phase] = total_iters
with open(test_logging_file, 'a') as f:
f.write('Epoch {}/{}， {} average acc:{:.4f} average threshold:{:.4f}'
.format(epoch, args.epoch-1, phase, np.mean(ACCs) * 100, np.mean(threshold))+'\n')
f.close()
model.train()
time_elapsed = time.time() - start
print('Finally Best Accuracy: LFW: {:.4f} in iters: {}, CFP_FP: {:.4f} in iters: {} and AgeDB-30: {:.4f} in iters: {}'.format(
best_acc['LFW'], best_iters['LFW'], best_acc['CFP_FP'], best_iters['CFP_FP'], best_acc['AgeDB30'], best_iters['AgeDB30']))
print('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))