-
Notifications
You must be signed in to change notification settings - Fork 0
/
FPN_Dliated_30_loss_train.py
281 lines (181 loc) · 8.7 KB
/
FPN_Dliated_30_loss_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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
from argparse import ArgumentParser
import os
import h5py
import torch
from torch.optim import Adam, lr_scheduler
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import Dataset
import numpy as np
import random
from scipy import stats
import datetime
import pandas as pd
import sys
from tqdm import tqdm
from backbones.CNN3D import *
from backbones.data_loader import *
if __name__ == "__main__":
parser = ArgumentParser(description='"my model')
parser.add_argument("--seed", type=int, default=19920517)
parser.add_argument('--lr', type=float, default=0.001,
help='learning rate (default: 0.001)')
parser.add_argument('--batch_size', type=int, default=64,
help='input batch size for training (default: 64)')
parser.add_argument('--epochs', type=int, default=2000,
help='number of epochs to train (default: 2000)')
parser.add_argument('--model', default='VSFA', type=str,
help='model name (default: VSFA)')
parser.add_argument('--weight_decay', type=float, default=0,
help='weight decay (default: 0.0)')
args = parser.parse_args()
args.decay_interval = int(args.epochs/100)
args.decay_ratio = 0.8
torch.manual_seed(args.seed) #
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
np.random.seed(args.seed)
random.seed(args.seed)
torch.utils.backcompat.broadcast_warning.enabled = True
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3"
num_for_val = 20
epoch_start=0
features_dir = "/cfs/cfs-3cab91f9f/liuzhang/open_datasets/FPN_D_30_origin_k_1200/"
print("训练数据目录：",features_dir)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
features_dir2 = "/cfs/cfs-3cab91f9f/liuzhang/open_datasets/FPN_D_30_origin_k_1200/"
val_list = "/cfs/cfs-3cab91f9f/liuzhang/open_datasets/KoNViD-1k_val.txt"
with open(val_list,"r") as f:
all_data = f.readlines()
val_data = []
for data in all_data:
val = data.split("_")[0]
val_data.append(val)
# 训练数据集合
videos_pic1 = []
result = os.listdir(features_dir)
total_videos = len(result)
#print("总数据:",total_videos)
width = height=0
max_len = 8000
train_list,val_list,test_list =[],[],[]
best_acc =0
for i in range(total_videos):
tmp = result[i].split(".")[0]
#rint(tmp)
if tmp in val_data:
val_list.append(result[i])
continue
train_list.append(result[i])
print(train_list[0],val_list[0])
print("split data:train: {}, test: {}, val: {}".format(len(train_list),len(test_list),len(val_list)))
# sys.exit()
#print(train_list[0])
# # print(len(train_index))
# train_list = train_list[0:100]
# val_list = val_list[0:10]
train_dataset = FPN_Dliated_VQADataset(features_dir,train_list, max_len=240,scale = 4.05)
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=args.batch_size, shuffle=True,num_workers=1)
print("load train data success!")
# for i, (features, length, label,name) in enumerate(train_loader):
# print(features.shape,length.shape)
# break
val_dataset = FPN_Dliated_VQADataset(features_dir2, val_list, max_len=240,scale = 4.05)
val_loader = torch.utils.data.DataLoader(dataset=val_dataset,)
print("load val data success!")
model = FPN_Dliated_LOSS_ResNet3D(Bottleneck, [3, 4, 6, 3]).to(device) #
if not os.path.exists('models'):
os.makedirs('models')
trained_model_file = 'models'
if torch.cuda.device_count() > 1:
print("Using", torch.cuda.device_count(), "GPUs!")
model = nn.DataParallel(model)
optimizer = Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
# pretrained =0
# if pretrained:
# new_state_dict = {}
# path = "./models/3D_VSFA1_acc:0.43177764565992865.pth"
# checkpoint = torch.load(path)
# # model.load_state_dict(checkpoint)
# for k, v in checkpoint["model"].items():
# name =k # remove `module.`，表面从第7个key值字符取到最后一个字符，正好去掉了module.
# new_state_dict[name] = v #新字典的key值对应的value为一一对应的值。
# model.load_state_dict(new_state_dict)
# epoch_start = checkpoint["epoch"]
# optimizer.load_state_dict(checkpoint['optimizer'])
# print("pre_modle:{}".format(path))
Not_well_video ={}
criterion = nn.L1Loss() # MSELoss loss
scheduler = lr_scheduler.StepLR(optimizer, step_size=args.decay_interval, gamma=args.decay_ratio)
best_PLCC = 0
for epoch in range(epoch_start,args.epochs):
# Train
model.train()
L = 0
right_num = 0
total_num = 0
print("training epch:{},total epch:{}".format(epoch,args.epochs))
for i, (features,D_features, length, label,name) in enumerate(tqdm(train_loader)):
# print("features,",features.shape,D_features.shape)
# print("length",length)
features = features.to(device).float()
D_features = D_features.to(device).float()
label = label.to(device).float()
optimizer.zero_grad() #
outputs_all = model(features,D_features)
outputs = outputs_all[0]
# print(outputs.shape)
outputs= outputs.squeeze(1)
# print(outputs.shape,label.shape)
# # print(outputs,label)
# sys.exit()
loss = criterion(outputs_all[0].squeeze(1), label) \
+ 0.5 * (criterion(outputs_all[1].squeeze(1), label) + 0.3 * criterion(outputs_all[2].squeeze(1), label) + 0.2 * criterion(outputs_all[3].squeeze(1), label))
loss.backward()
optimizer.step()
L = L + loss.item()
train_loss = L / (i + 1)
print("train_loss:",train_loss)
if epoch % num_for_val ==0:
print("start valling")
model.eval()
# Val
y_pred = np.zeros(len(val_list))
y_val = np.zeros(len(val_list))
L = 0
with torch.no_grad():
badcase = {}
# y_pred = np.zeros(len(result))
# y_test = np.zeros(len(result))
L = 0
for i, (features,D_features, length, label,name) in enumerate(tqdm(val_loader)):
y_val[i] = 4.05*label.item()
features = features.to(device).float()
D_features = D_features.to(device).float()
label = label.to(device).float()
outputs_all = model(features,D_features)
outputs = outputs_all[0]
y_pred[i] = 4.05*outputs.item()
#y_pred[i] = 1 * outputs.item()
loss = criterion(outputs, label)
#print(outputs,label)
L = L + loss.item()
val_loss = L / (i + 1)
val_PLCC = stats.pearsonr(y_pred, y_val)[0]
val_SROCC = stats.spearmanr(y_pred, y_val)[0]
val_RMSE = np.sqrt(((y_pred-y_val) ** 2).mean())
val_KROCC = stats.stats.kendalltau(y_pred, y_val)[0]
#print("Badcase",badcase)
print("Val results: val loss={:.4f}, SROCC={:.4f}, KROCC={:.4f}, PLCC={:.4f}, RMSE={:.4f}" .format(val_loss, val_SROCC, val_KROCC, val_PLCC, val_RMSE))
if val_PLCC >best_PLCC:
print("save model at epch")
state = {'model': model.state_dict(), 'optimizer': optimizer.state_dict(), 'epoch': epoch}
torch.save(state,os.path.join(trained_model_file,"New_loss222_{}_plcc:{}.pth".format(epoch+1,val_PLCC)))
print("Epoch {} model saved!".format(epoch + 1))
best_PLCC = val_PLCC
elif epoch % 20 ==0:
print("save model at epch")
state = {'model': model.state_dict(), 'optimizer': optimizer.state_dict(), 'epoch': epoch}
torch.save(state,os.path.join(trained_model_file,"New_loss222_{}_epoch:{}.pth".format(epoch+1,epoch+1)))
print("Epoch {} model saved!".format(epoch + 1))
# print(badcase)