-
Notifications
You must be signed in to change notification settings - Fork 18
/
train.py
144 lines (128 loc) · 7.54 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
import argparse
import pandas as pd
import torch
from thop import profile, clever_format
from torch.optim import Adam
from torch.optim.lr_scheduler import MultiStepLR
from torch.utils.data import DataLoader
from tqdm import tqdm
from model import Model, set_bn_eval
from utils import recall, LabelSmoothingCrossEntropyLoss, BatchHardTripletLoss, ImageReader, MPerClassSampler
def train(net, optim):
net.train()
# fix bn on backbone network
net.apply(set_bn_eval)
total_loss, total_correct, total_num, data_bar = 0, 0, 0, tqdm(train_data_loader)
for inputs, labels in data_bar:
inputs, labels = inputs.cuda(), labels.cuda()
features, classes = net(inputs)
class_loss = class_criterion(classes, labels)
feature_loss = feature_criterion(features, labels)
loss = class_loss + feature_loss
optim.zero_grad()
loss.backward()
optim.step()
pred = torch.argmax(classes, dim=-1)
total_loss += loss.item() * inputs.size(0)
total_correct += torch.sum(pred == labels).item()
total_num += inputs.size(0)
data_bar.set_description('Train Epoch {}/{} - Loss:{:.4f} - Acc:{:.2f}%'
.format(epoch, num_epochs, total_loss / total_num, total_correct / total_num * 100))
return total_loss / total_num, total_correct / total_num * 100
def test(net, recall_ids):
net.eval()
with torch.no_grad():
# obtain feature vectors for all data
for key in eval_dict.keys():
eval_dict[key]['features'] = []
for inputs, labels in tqdm(eval_dict[key]['data_loader'], desc='processing {} data'.format(key)):
inputs, labels = inputs.cuda(), labels.cuda()
features, classes = net(inputs)
eval_dict[key]['features'].append(features)
eval_dict[key]['features'] = torch.cat(eval_dict[key]['features'], dim=0)
# compute recall metric
if data_name == 'isc':
acc_list = recall(eval_dict['test']['features'], test_data_set.labels, recall_ids,
eval_dict['gallery']['features'], gallery_data_set.labels)
else:
acc_list = recall(eval_dict['test']['features'], test_data_set.labels, recall_ids)
desc = 'Test Epoch {}/{} '.format(epoch, num_epochs)
for index, rank_id in enumerate(recall_ids):
desc += 'R@{}:{:.2f}% '.format(rank_id, acc_list[index] * 100)
results['test_recall@{}'.format(rank_id)].append(acc_list[index] * 100)
print(desc)
return acc_list[0]
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Train CGD')
parser.add_argument('--data_path', default='/home/data', type=str, help='datasets path')
parser.add_argument('--data_name', default='car', type=str, choices=['car', 'cub', 'sop', 'isc'],
help='dataset name')
parser.add_argument('--crop_type', default='uncropped', type=str, choices=['uncropped', 'cropped'],
help='crop data or not, it only works for car or cub dataset')
parser.add_argument('--backbone_type', default='resnet50', type=str, choices=['resnet50', 'resnext50'],
help='backbone network type')
parser.add_argument('--gd_config', default='SG', type=str,
choices=['S', 'M', 'G', 'SM', 'MS', 'SG', 'GS', 'MG', 'GM', 'SMG', 'MSG', 'GSM'],
help='global descriptors config')
parser.add_argument('--feature_dim', default=1536, type=int, help='feature dim')
parser.add_argument('--smoothing', default=0.1, type=float, help='smoothing value for label smoothing')
parser.add_argument('--temperature', default=0.5, type=float,
help='temperature scaling used in softmax cross-entropy loss')
parser.add_argument('--margin', default=0.1, type=float, help='margin of m for triplet loss')
parser.add_argument('--recalls', default='1,2,4,8', type=str, help='selected recall')
parser.add_argument('--batch_size', default=128, type=int, help='train batch size')
parser.add_argument('--num_epochs', default=20, type=int, help='train epoch number')
opt = parser.parse_args()
# args parse
data_path, data_name, crop_type, backbone_type = opt.data_path, opt.data_name, opt.crop_type, opt.backbone_type
gd_config, feature_dim, smoothing, temperature = opt.gd_config, opt.feature_dim, opt.smoothing, opt.temperature
margin, recalls, batch_size = opt.margin, [int(k) for k in opt.recalls.split(',')], opt.batch_size
num_epochs = opt.num_epochs
save_name_pre = '{}_{}_{}_{}_{}_{}_{}_{}_{}'.format(data_name, crop_type, backbone_type, gd_config, feature_dim,
smoothing, temperature, margin, batch_size)
results = {'train_loss': [], 'train_accuracy': []}
for recall_id in recalls:
results['test_recall@{}'.format(recall_id)] = []
# dataset loader
train_data_set = ImageReader(data_path, data_name, 'train', crop_type)
train_sample = MPerClassSampler(train_data_set.labels, batch_size)
train_data_loader = DataLoader(train_data_set, batch_sampler=train_sample, num_workers=8)
test_data_set = ImageReader(data_path, data_name, 'query' if data_name == 'isc' else 'test', crop_type)
test_data_loader = DataLoader(test_data_set, batch_size, shuffle=False, num_workers=8)
eval_dict = {'test': {'data_loader': test_data_loader}}
if data_name == 'isc':
gallery_data_set = ImageReader(data_path, data_name, 'gallery', crop_type)
gallery_data_loader = DataLoader(gallery_data_set, batch_size, shuffle=False, num_workers=8)
eval_dict['gallery'] = {'data_loader': gallery_data_loader}
# model setup, model profile, optimizer config and loss definition
model = Model(backbone_type, gd_config, feature_dim, num_classes=len(train_data_set.class_to_idx)).cuda()
flops, params = profile(model, inputs=(torch.randn(1, 3, 224, 224).cuda(),))
flops, params = clever_format([flops, params])
print('# Model Params: {} FLOPs: {}'.format(params, flops))
optimizer = Adam(model.parameters(), lr=1e-4)
lr_scheduler = MultiStepLR(optimizer, milestones=[int(0.6 * num_epochs), int(0.8 * num_epochs)], gamma=0.1)
class_criterion = LabelSmoothingCrossEntropyLoss(smoothing=smoothing, temperature=temperature)
feature_criterion = BatchHardTripletLoss(margin=margin)
best_recall = 0.0
for epoch in range(1, num_epochs + 1):
train_loss, train_accuracy = train(model, optimizer)
results['train_loss'].append(train_loss)
results['train_accuracy'].append(train_accuracy)
rank = test(model, recalls)
lr_scheduler.step()
# save statistics
data_frame = pd.DataFrame(data=results, index=range(1, epoch + 1))
data_frame.to_csv('results/{}_statistics.csv'.format(save_name_pre), index_label='epoch')
# save database and model
data_base = {}
if rank > best_recall:
best_recall = rank
data_base['test_images'] = test_data_set.images
data_base['test_labels'] = test_data_set.labels
data_base['test_features'] = eval_dict['test']['features']
if data_name == 'isc':
data_base['gallery_images'] = gallery_data_set.images
data_base['gallery_labels'] = gallery_data_set.labels
data_base['gallery_features'] = eval_dict['gallery']['features']
torch.save(model.state_dict(), 'results/{}_model.pth'.format(save_name_pre))
torch.save(data_base, 'results/{}_data_base.pth'.format(save_name_pre))