-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
213 lines (166 loc) · 9.27 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
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
import os
import argparse
import yaml
from tqdm import tqdm
import numpy as np
import random
import time
import torch
from torch.utils.data import DataLoader
#from tensorboardX import SummaryWriter
import train_utils
from train_utils import AverageMeter
from eval_utils import eval, cal_rmse
from dataset import SHT_LPN_Train_dataset, SHT_LPN_Test_dataset, UCF_Dataset, XD_Dataset
from model.RAPN import RAPN
from losses import Total_loss
def eval_epoch(config, model, test_dataloader):
model = model.eval()
total_labels, total_scores = [], []
for frames in tqdm(test_dataloader):
frames = frames.view(-1, frames.shape[-2], frames.shape[-1])
# frames=frames.float().contiguous().view([-1, 1, frames.shape[-1]]).cuda()
# flows=flows.float().contiguous().view([-1, 1, flows.shape[-1]]).cuda()
with torch.no_grad():
score = model(frames, frames, isTrain=False, gl=config['test_gl'])
if config['dataset_name'] == 'UCF':
score = torch.mean(score, dim=0)
score = list(score.cpu().detach().numpy())
score = np.repeat(np.array(score), 16)
total_scores.extend(score)
if config['dataset_name'] == 'XDViolence':
total_scores = np.array(total_scores).reshape([-1, 5])
total_scores = np.mean(total_scores, axis=1)
gt = list(np.load(config['gt']))
return eval(total_scores, gt)
def train(config):
#### set the save and log path ####
save_path = config['save_path']
train_utils.set_save_path(save_path)
train_utils.set_log_path(save_path)
#writer = SummaryWriter(os.path.join(config['save_path'], 'tensorboard'))
yaml.dump(config, open(os.path.join(config['save_path'], 'classifier_config.yaml'), 'w'))
#### make datasets ####
def worker_init(worked_id):
np.random.seed(worked_id)
random.seed(worked_id)
# train
if config['dataset_name'] == 'SHT':
ref_dataset_nor = SHT_LPN_Train_dataset(config['rgb_dataset_path'], config['train_split'],
config['clips_num'], segment_len=config['segment_len'], type='Normal', ten_crop=config['ten_crop'], real_label=False)
ref_dataset_abn = SHT_LPN_Train_dataset(config['rgb_dataset_path'],config['train_split'],
config['clips_num'],segment_len=config['segment_len'], type='Abnormal', ten_crop=config['ten_crop'], real_label=False)
test_dataset = SHT_LPN_Test_dataset(config['rgb_dataset_path'], config['test_split'],
config['test_mask_dir'], segment_len=config['segment_len'],
ten_crop=config['ten_crop'])
elif config['dataset_name'] == 'UCF':
ref_dataset_nor = UCF_Dataset(config['rgb_dataset_list'], config['test_rgb_dataset_list'],
clip_num=config['clips_num'], is_normal=True)
ref_dataset_abn = UCF_Dataset(config['rgb_dataset_list'], config['test_rgb_dataset_list'],
clip_num=config['clips_num'], is_normal=False)
test_dataset = UCF_Dataset(config['rgb_dataset_list'], config['test_rgb_dataset_list'],
clip_num=config['clips_num'], test_mode=True)
elif config['dataset_name'] == 'XDViolence':
ref_dataset_nor = XD_Dataset(config['rgb_dataset_list'], config['test_rgb_dataset_list'], clip_num=config['clips_num'], is_normal=True)
ref_dataset_abn = XD_Dataset(config['rgb_dataset_list'], config['test_rgb_dataset_list'], clip_num=config['clips_num'], is_normal=False)
test_dataset = XD_Dataset(config['rgb_dataset_list'], config['test_rgb_dataset_list'], clip_num=config['clips_num'], test_mode=True)
ref_dataloader_nor = DataLoader(ref_dataset_nor, batch_size=config['batch_size'], shuffle=True,
num_workers=16, worker_init_fn=worker_init, drop_last=True, pin_memory=True)
ref_dataloader_abn = DataLoader(ref_dataset_abn, batch_size=config['batch_size'], shuffle=True,
num_workers=16, worker_init_fn=worker_init, drop_last=True, pin_memory=True)
# test
test_dataloader = DataLoader(test_dataset, batch_size=config['test_batch_size'], shuffle=False,
num_workers=8, worker_init_fn=worker_init, drop_last=False, pin_memory=True)
#### Model setting ####
model = RAPN(config['feature_dim'], config['out_feature_dim'], config['layer_num'], config['dropout_rate']).cuda()
# optimizer setting
params = list(model.parameters())
optimizer, lr_scheduler = train_utils.make_optimizer(
params, config['optimizer'], config['optimizer_args'])
model = torch.nn.parallel.DataParallel(model, dim=0, device_ids=config['gpu'])
model = model.train()
criterion = Total_loss(config).cuda()
train_utils.log('Start train')
iterator = 0
test_epoch = 10 if config['eval_epoch'] is None else config['eval_epoch']
AUCs,tious,best_epoch,best_tiou_epoch,best_tiou,best_AUC=[],[],0,0,0,0
ref_iter_abn = iter(ref_dataloader_abn)
auc = eval_epoch(config, model, test_dataloader)
print(auc)
for epoch in range(config['epochs']):
model = model.train()
Errs, Atten_Errs, Rmses = AverageMeter(), AverageMeter(), AverageMeter()
for step, (ref_rgb_nor, ref_labels_nor) in tqdm(enumerate(ref_dataloader_nor)):
try:
ref_rgb_abn, ref_labels_abn = next(ref_iter_abn)
except:
del ref_iter_abn
ref_iter_abn = iter(ref_dataloader_abn)
ref_rgb_abn, ref_labels_abn = next(ref_iter_abn)
ref_rgb_nor, ref_labels_nor= ref_rgb_nor.cuda().float(), ref_labels_nor.cuda().float()
ref_rgb_abn, ref_labels_abn = ref_rgb_abn.cuda().float(), ref_labels_abn.cuda().float()
ref_rgb_nor = ref_rgb_nor.view(-1, ref_rgb_nor.shape[-2], ref_rgb_nor.shape[-1])
ref_rgb_abn = ref_rgb_abn.view(-1, ref_rgb_abn.shape[-2], ref_rgb_abn.shape[-1])
p_score, label, abn_score_topK_nor, abn_score_topK_abn, nor_feat_conf_nor, abn_feat_conf_nor, abn_feat_conf_abn \
= model(ref_rgb_nor, ref_rgb_abn)
p_label = torch.cat([ref_labels_nor, ref_labels_abn], dim=0)
cost = criterion(p_score, p_label, label, abn_score_topK_nor, abn_score_topK_abn, nor_feat_conf_nor, abn_feat_conf_nor, abn_feat_conf_abn)
cost.backward()
optimizer.step()
optimizer.zero_grad()
Errs.update(cost)
iterator += 1
# if iterator % 10 == 0 and epoch > 14:
#
# auc = eval_epoch(config, model, test_dataloader)
# AUCs.append(auc)
# if len(AUCs) >= 5:
# mean_auc = sum(AUCs[-5:]) / 5.
# if mean_auc > best_AUC:
# best_epoch, best_AUC = epoch, mean_auc
# train_utils.log('best_AUC {} at epoch {}, now {}'.format(best_AUC, best_epoch, mean_auc))
#
# train_utils.log('===================')
# if auc > 0.84:
# checkpoint = {
# 'model': model.state_dict(),
# 'optimizer': optimizer.state_dict(),
# }
# torch.save(checkpoint,
# os.path.join(save_path, 'models/model-epoch-{}-AUC-{}.pth'.format(epoch, auc)))
train_utils.log('[{}]: err\t{:.4f}\tatten\t{:.4f}'.format(epoch, Errs, Atten_Errs))
Errs.reset(), Atten_Errs.reset()
train_utils.log("epoch {}, lr {}".format(epoch, optimizer.param_groups[0]['lr']))
train_utils.log('----------------------------------------')
if epoch % test_epoch == 0 and epoch > 4:
auc = eval_epoch(config, model, test_dataloader)
AUCs.append(auc)
if len(AUCs) >= 5:
mean_auc = sum(AUCs[-5:]) / 5.
if mean_auc > best_AUC:
best_epoch,best_AUC =epoch,mean_auc
train_utils.log('best_AUC {} at epoch {}, now {}'.format(best_AUC, best_epoch, mean_auc))
train_utils.log('===================')
if auc > 0.8:
checkpoint = {
'model': model.state_dict(),
'optimizer': optimizer.state_dict(),
}
torch.save(checkpoint, os.path.join(save_path, 'models/model-epoch-{}-AUC-{}.pth'.format(epoch, auc)))
train_utils.log('Training is finished')
train_utils.log('max_frame_AUC: {}'.format(best_AUC))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--config')
parser.add_argument('--tag', default=None)
parser.add_argument('--gpu', default='0')
# for flownet2, no need to modify
parser.add_argument('--fp16', action='store_true', help='Run model in pseudo-fp16 mode (fp16 storage fp32 math).')
parser.add_argument("--rgb_max", type=float, default=255.)
args = parser.parse_args()
config = yaml.load(open(args.config, 'r'), Loader=yaml.FullLoader)
if args.tag is not None:
config['save_path'] += ('_' + args.tag)
train_utils.set_gpu(args.gpu)
config['gpu'] =[i for i in range(len(args.gpu.split(',')))]
train(config)