-
Notifications
You must be signed in to change notification settings - Fork 5
/
train_prcc_pcb.py
135 lines (108 loc) · 5.57 KB
/
train_prcc_pcb.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
# encoding: utf-8
import argparse
import os
import sys
import torch
from torch.backends import cudnn
sys.path.append('.')
from data import make_data_loader_prcc_hrnet as make_data_loader
from engine.trainer import do_train_prcc_pcb as do_train
from modeling import build_model
from layers import make_loss_with_pcb
from solver import make_optimizer_with_pcb
from solver import WarmupMultiStepLR
from engine.inference import inference_prcc_pcb
import datetime
def load_network_pretrain(model, cfg):
path = os.path.join(cfg.logs_dir, 'checkpoint_best.pth')
if not os.path.exists(path):
return model, 0, 0.0
pre_dict = torch.load(path)
model.load_state_dict(pre_dict['state_dict'])
start_epoch = pre_dict['epoch']
best_acc = pre_dict['best_acc']
print('start_epoch:', start_epoch)
print('best_acc:', best_acc)
return model, start_epoch, best_acc
def main(cfg):
# prepare dataset
train_loader, train_loader_ca, train_loader_cb, val_loader_c, val_loader_b, num_query_c, num_query_b, num_classes = make_data_loader(cfg, h=256, w=128) # num_query=3368, num_classes=751
# prepare model
model = build_model(num_classes, 'pcb') # num_classes=751
model = torch.nn.DataParallel(model).cuda() if torch.cuda.is_available() else model
loss_func = make_loss_with_pcb(cfg, num_classes)
optimizer = make_optimizer_with_pcb(cfg, model)
if cfg.lr_type == 'step':
scheduler = WarmupMultiStepLR(optimizer, cfg.steps, cfg.gamma, cfg.warmup_factor, cfg.warmup_iters, cfg.warmup_method)
else:
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'max', patience=10, factor=0.5)
if cfg.train == 'train':
start_epoch = int(0)
acc_best = 0.0
if cfg.resume == 1:
model, start_epoch, acc_best = load_network_pretrain(model, cfg)
do_train(cfg, model, train_loader, val_loader_c, optimizer, scheduler, loss_func, num_query_c, start_epoch, acc_best, lr_type=cfg.lr_type)
elif cfg.train == 'test':
# Test
last_model_wts = torch.load(os.path.join(cfg.logs_dir, 'checkpoint_best.pth'))
model.load_state_dict(last_model_wts['state_dict'])
mAP, cmc1 = inference_prcc_pcb(model, val_loader_c, num_query_c)
start_time = datetime.datetime.now()
start_time = '%4d:%d:%d-%2d:%2d:%2d' % (start_time.year, start_time.month, start_time.day, start_time.hour, start_time.minute, start_time.second)
line = '{} - Test: cmc1: {:.1%}, mAP: {:.1%}\n'.format(start_time, cmc1, mAP)
print(line)
if __name__ == '__main__':
gpu_id = 0
os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_id)
cudnn.benchmark = True
parser = argparse.ArgumentParser(description="ReID Baseline Training")
# DATA
parser.add_argument('--batch_size', type=int, default=64)
parser.add_argument('--img_per_id', type=int, default=4)
parser.add_argument('--batch_size_test', type=int, default=128)
parser.add_argument('--workers', type=int, default=8)
parser.add_argument('--height', type=int, default=256)
parser.add_argument('--width', type=int, default=128)
parser.add_argument('--height_mask', type=int, default=256)
parser.add_argument('--width_mask', type=int, default=128)
# MODEL
parser.add_argument('--features', type=int, default=128)
parser.add_argument('--dropout', type=float, default=0.0)
parser.add_argument('--parts', type=int, default=6)
# OPTIMIZER
parser.add_argument('--seed', type=int, default=1)
parser.add_argument('--lr', type=float, default=0.0035)
parser.add_argument('--lr_center', type=float, default=0.5)
parser.add_argument('--lr_type', type=str, default='step', help='step, plateau')
parser.add_argument('--center_loss_weight', type=float, default=0.0005)
parser.add_argument('--steps', type=list, default=[40, 80])
parser.add_argument('--gamma', type=float, default=0.1)
parser.add_argument('--cluster_margin', type=float, default=0.3)
parser.add_argument('--bias_lr_factor', type=float, default=1.0)
parser.add_argument('--weight_decay', type=float, default=5e-4)
parser.add_argument('--weight_decay_bias', type=float, default=5e-4)
parser.add_argument('--range_k', type=float, default=2)
parser.add_argument('--range_margin', type=float, default=0.3)
parser.add_argument('--range_alpha', type=float, default=0)
parser.add_argument('--range_beta', type=float, default=1)
parser.add_argument('--range_loss_weight', type=float, default=1)
parser.add_argument('--warmup_factor', type=float, default=0.01)
parser.add_argument('--warmup_iters', type=float, default=10)
parser.add_argument('--warmup_method', type=str, default='linear')
parser.add_argument('--margin', type=float, default=0.3)
parser.add_argument('--optimizer_name', type=str, default="SGD")
parser.add_argument('--momentum', type=float, default=0.9)
# TRAINER
parser.add_argument('--max_epochs', type=int, default=60)
parser.add_argument('--train', type=str, default='train', help='train, test') # change train or test mode
parser.add_argument('--resume', type=int, default=0)
parser.add_argument('--num_works', type=int, default=8)
# misc
working_dir = os.path.dirname(os.path.abspath(__file__))
parser.add_argument('--dataset', type=str, default='prcc_gcn')
parser.add_argument('--data_dir', type=str, default='/data/prcc/')
parser.add_argument('--logs_dir', type=str, default=os.path.join(working_dir, 'logs/prcc_pcb/'))
cfg = parser.parse_args()
if not os.path.exists(cfg.logs_dir):
os.makedirs(cfg.logs_dir)
main(cfg)