-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_source.py
174 lines (135 loc) · 7.2 KB
/
train_source.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
import os
import shutil
import torch
import numpy as np
from tqdm import tqdm
from model.SFUniDA import SFUniDA
from dataset.dataset import SFUniDADataset
from torch.utils.data.dataloader import DataLoader
from config.model_config import build_args
from utils.net_utils import set_logger, set_random_seed
from utils.net_utils import compute_h_score, CrossEntropyLabelSmooth
def op_copy(optimizer):
for param_group in optimizer.param_groups:
param_group['lr0'] = param_group['lr']
return optimizer
def lr_scheduler(optimizer, iter_num, max_iter, gamma=10, power=0.75):
decay = (1 + gamma * iter_num / max_iter) ** (-power)
for param_group in optimizer.param_groups:
param_group['lr'] = param_group['lr0'] * decay
param_group['weight_decay'] = 1e-3
param_group['momentum'] = 0.9
param_group['nesterov'] = True
return optimizer
def train(args, model, dataloader, criterion, optimizer, epoch_idx=0.0):
model.train()
loss_stack = []
iter_idx = epoch_idx * len(dataloader)
iter_max = args.epochs * len(dataloader)
for imgs_train, _, imgs_label, _ in tqdm(dataloader, ncols=60):
iter_idx += 1
imgs_train = imgs_train.cuda()
imgs_label = imgs_label.cuda()
_, pred_cls = model(imgs_train, apply_softmax=True)
imgs_onehot_label = torch.zeros_like(pred_cls).scatter(1, imgs_label.unsqueeze(1), 1)
loss = criterion(pred_cls, imgs_onehot_label)
lr_scheduler(optimizer, iter_idx, iter_max)
optimizer.zero_grad()
loss.backward()
optimizer.step()
loss_stack.append(loss.cpu().item())
train_loss = np.mean(loss_stack)
return train_loss
@torch.no_grad()
def test(args, model, dataloader, src_flg=True):
model.eval()
gt_label_stack = []
pred_cls_stack = []
if src_flg:
class_list = args.source_class_list
open_flg = False
else:
class_list = args.target_class_list
open_flg = args.target_private_class_num > 0
for _, imgs_test, imgs_label, _ in tqdm(dataloader, ncols=60):
imgs_test = imgs_test.cuda()
_, pred_cls = model(imgs_test, apply_softmax=True)
gt_label_stack.append(imgs_label)
pred_cls_stack.append(pred_cls.cpu())
gt_label_all = torch.cat(gt_label_stack, dim=0) #[N]
pred_cls_all = torch.cat(pred_cls_stack, dim=0) #[N, C]
h_score, known_acc,\
unknown_acc, per_cls_acc = compute_h_score(args, class_list, gt_label_all, pred_cls_all, open_flg, open_thresh=0.50)
return h_score, known_acc, unknown_acc, per_cls_acc
def main(args):
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
this_dir = os.path.join(os.path.dirname(__file__), ".")
model = SFUniDA(args)
if args.checkpoint is not None and os.path.isfile(args.checkpoint):
save_dir = os.path.dirname(args.checkpoint)
checkpoint = torch.load(args.checkpoint, map_location=torch.device("cpu"))
model.load_state_dict(checkpoint["model_state_dict"])
else:
save_dir = os.path.join(this_dir, "checkpoints_glc", args.dataset, "source_{}".format(args.s_idx),
"source_{}_{}".format(args.source_train_type, args.target_label_type))
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
model.cuda()
args.save_dir = save_dir
logger = set_logger(args, log_name="log_source_training.txt")
params_group = []
for k, v in model.backbone_layer.named_parameters():
params_group += [{"params":v, 'lr':args.lr*0.1}]
for k, v in model.feat_embed_layer.named_parameters():
params_group += [{"params":v, 'lr':args.lr}]
for k, v in model.class_layer.named_parameters():
params_group += [{"params":v, 'lr':args.lr}]
optimizer = torch.optim.SGD(params_group)
optimizer = op_copy(optimizer)
source_data_list = open(os.path.join(args.source_data_dir, "image_unida_list.txt"), "r").readlines()
source_dataset = SFUniDADataset(args, args.source_data_dir, source_data_list, d_type="source", preload_flg=True)
source_dataloader = DataLoader(source_dataset, batch_size=args.batch_size, shuffle=True,
num_workers=args.num_workers, drop_last=True)
target_dataloader_list = []
for idx in range(len(args.target_domain_dir_list)):
target_data_dir = args.target_domain_dir_list[idx]
target_data_list = open(os.path.join(target_data_dir, "image_unida_list.txt"), "r").readlines()
target_dataset = SFUniDADataset(args, target_data_dir, target_data_list, d_type="target", preload_flg=False)
target_dataloader_list.append(DataLoader(target_dataset, batch_size=args.batch_size, shuffle=False,
num_workers=args.num_workers, drop_last=False))
if args.source_train_type == "smooth":
criterion = CrossEntropyLabelSmooth(num_classes=args.class_num, epsilon=0.1, reduction=True)
elif args.source_train_type == "vanilla":
criterion = CrossEntropyLabelSmooth(num_classes=args.class_num, epsilon=0.0, reduction=True)
else:
raise ValueError("Unknown source_train_type:", args.source_train_type)
notation_str = "\n=================================================\n"
notation_str += " START TRAINING ON THE SOURCE:{} == {} \n".format(args.s_idx, args.target_label_type)
notation_str += "================================================="
logger.info(notation_str)
for epoch_idx in tqdm(range(args.epochs), ncols=60):
train_loss = train(args, model, source_dataloader, criterion, optimizer, epoch_idx)
logger.info("Epoch:{}/{} train_loss:{:.3f}".format(epoch_idx, args.epochs, train_loss))
if epoch_idx % 1 == 0:
# EVALUATE ON SOURCE
source_h_score, source_known_acc, source_unknown_acc, src_per_cls_acc = test(args, model, source_dataloader, src_flg=True)
logger.info("EVALUATE ON SOURCE: H-Score:{:.3f}, KnownAcc:{:.3f}, UnknownAcc:{:.3f}".\
format(source_h_score, source_known_acc, source_unknown_acc))
if args.dataset == "VisDA":
logger.info("VISDA PER_CLS_ACC:")
logger.info(src_per_cls_acc)
checkpoint_file = "latest_source_checkpoint.pth"
torch.save({
"epoch":epoch_idx,
"model_state_dict":model.state_dict()}, os.path.join(save_dir, checkpoint_file))
for idx_i, item in enumerate(args.target_domain_list):
notation_str = "\n=================================================\n"
notation_str += " EVALUATE ON THE TARGET:{} \n".format(item)
notation_str += "================================================="
logger.info(notation_str)
hscore, knownacc, unknownacc, _ = test(args, model, target_dataloader_list[idx_i], src_flg=False)
logger.info("H-Score:{:.3f}, KnownAcc:{:.3f}, UnknownACC:{:.3f}".format(hscore, knownacc, unknownacc))
if __name__ == "__main__":
args = build_args()
set_random_seed(args.seed)
main(args)