-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathtrain_style_classification.py
99 lines (73 loc) · 3.14 KB
/
train_style_classification.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
import os
from pprint import pformat
import pytorch_lightning as pl
import torch
from omegaconf import OmegaConf
from pytorch_lightning import loggers as pl_loggers
from pytorch_lightning.callbacks import ModelCheckpoint
from statistics import mean
from torch.optim import AdamW
import torch.nn as nn
from mld.callback import ProgressLogger
from mld.config import parse_args
from mld.data.get_data import get_datasets
from mld.models.get_model import get_model
from mld.utils.logger import create_logger
from mld.models.architectures.mld_style_encoder import StyleClassification
from mld.data.humanml.data.dataset import StyleMotionDataset
def reset_loss_dict(loss_dict):
loss_dict = {
"crossentropy_loss": []
}
def main():
# create dataset
datasets = StyleMotionDataset('train')
test_dataset = StyleMotionDataset('test')
dataloader = torch.utils.data.DataLoader(dataset=datasets, batch_size=128, shuffle=True,drop_last=True)
testloader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=16, shuffle=True)
criterion = torch.nn.CrossEntropyLoss()
model = StyleClassification(nclasses=47).cuda()
optimizer = AdamW(params=filter(lambda p: p.requires_grad, model.parameters()), lr=1e-4)
loss_dict = {
"crossentropy_loss": [],
"tri_loss":[]
}
log_freq = 2
n_epoch = 200
test_freq = 10
save_freq = 100
model_dir = "./experiments"
model.train()
for epoch in range(n_epoch):
total = 0
correct = 0
for i, train_data in enumerate(dataloader):
optimizer.zero_grad()
motion_input = train_data['motion']
output_score,feat = model(motion_input,stage="Both")
_, predicted = torch.max(output_score, 1)
total += train_data["motion"].size(0)
correct += (predicted == train_data["label"]).sum().item()
loss = criterion(output_score,train_data['label']) +0.001*center_loss(feat,train_data['label'])
loss_all = loss
loss_dict["crossentropy_loss"].append(loss.item())
loss_all.backward()
optimizer.step()
if (i + 1) % log_freq == 0:
print('Train: Epoch [{}/{}], Step [{}/{}]| loss: {:.4f} accuracy: {:.4f}'.format(epoch + 1, n_epoch, i + 1, len(dataloader), mean(loss_dict["crossentropy_loss"]),100*(correct / total)))
if (epoch + 1) % test_freq == 0:
model.eval()
total = 0
correct = 0
for i, test_data in enumerate(testloader):
output_score = model(test_data['motion'])
_, predicted = torch.max(output_score, 1)
total += test_data["motion"].size(0)
correct += (predicted == test_data["label"]).sum().item()
print('Test: Epoch [{}/{}]| accuracy: {:.4f}%'
.format(epoch + 1, n_epoch, 100*(correct / total)))
model.train()
if (epoch+1) % save_freq == 0:
torch.save(model.state_dict(), os.path.join(model_dir, "style_encoder_robust47_{}.pt".format(epoch+1)))
if __name__ == "__main__":
main()