-
Notifications
You must be signed in to change notification settings - Fork 3
/
train.py
68 lines (57 loc) · 2.12 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
import torch
from torch import nn
from utils import to_var, batch_ids2words
from build_vocab import Vocabulary
from pycocotools.coco import COCO
from PIL import Image
from loss import caption_loss, class_loss
def train(args,
vocab,
train_loader,
model,
caption_criterion,
classes_criterion,
mode,
optimizer,
loss_type,
epoch):
model.train()
total_num = 0
loss_caption_total = 0.0
loss_class_total = 0.0
total_step = len(train_loader)
for i, (images, classes, captions) in enumerate(train_loader):
# Set mini-batch dataset
images = to_var(images)
captions = to_var(captions)
classes = to_var(classes)
batch_size = images.size(0)
# Forward, Backward and Optimize
optimizer.zero_grad()
outputs_captions, outputs_classes = model(images, captions[::, :-1])
# classes loss
loss_class = class_loss(loss_type, outputs_classes, classes, classes_criterion)
# caption loss
loss_caption = caption_loss(outputs_captions, captions[::, 1:], caption_criterion)
if mode == 'none':
loss = loss_caption
elif mode == 'class_only':
loss = loss_class
elif mode == 'caption_only':
loss = loss_caption
else:
raise IndexError('No such mode')
loss.backward()
optimizer.step()
loss_caption_total += loss_caption.item()
loss_class_total += loss_class.item()
# calculate the metric scores
# references = batch_ids2words(captions.view(1, -1), vocab)
# candidates = batch_ids2words(output_captions.view(1, -1), vocab)
# classes = batch_ids2words(output_classes.view(1, -1), vocab)
# Print log info
total_num += 1
if (i%20==0 and i!=0) or i==len(train_loader)-1:
print('Epoch [%d/%d], Step [%d/%d], Caption Loss: %.4f, Class Loss: %5.4f'
%(epoch, args.end_epoch, i, total_step,
loss_caption_total/total_num, loss_class_total/total_num))