-
Notifications
You must be signed in to change notification settings - Fork 5
/
main.py
153 lines (132 loc) · 6.82 KB
/
main.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
import torch
import torch.nn as nn
from models.Net import AlexNet, Uniform_D
from dataset.customData import MyCustomDataset
from loss.contrast import Contrast_Loss, Quantization_Loss
import numpy as np
from torch.autograd import Variable
from torchvision import transforms
from torch.utils.data import DataLoader
from utils.LoadWeights import load_preweights
from utils.generateUniformData import generate_binary_distribution
from tqdm import tqdm
import os
import argparse
torch.cuda.manual_seed_all(1)
np.random.seed(1)
def calculate_classification_accuracy(predict, target):
predict_label = torch.argmax(predict.data, 1)
correct_pred = (predict_label == target.data).sum().item()
return correct_pred
def train(args):
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpus
transform = transforms.Compose([
transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
trainset = MyCustomDataset(root_path=args.img_tr, transform=transform)
testset = MyCustomDataset(root_path=args.img_te, transform=transform)
trainloader = DataLoader(trainset,
batch_size=args.batchsize,
shuffle=True,
drop_last=True)
testloader = DataLoader(testset, batch_size=args.batchsize, shuffle=True)
G = AlexNet(num_classes=args.label_dim, Kbits=args.Kbits)
G = G.cuda().float()
state_dict = load_preweights(G, args.initialized)
G.load_state_dict(state_dict)
crossentropy = nn.CrossEntropyLoss()
optimizer_G = torch.optim.Adam(G.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
# Adversarial ground truths
valid = Variable(torch.Tensor(args.batchsize, 1).fill_(1.0),
requires_grad=False).cuda()
fake = Variable(torch.Tensor(args.batchsize, 1).fill_(0.0),
requires_grad=False).cuda()
D = Uniform_D(Kbits=args.Kbits)
D = D.cuda().float()
optimizer_D = torch.optim.Adam(D.parameters(), lr=args.lr)
adversarial_loss = torch.nn.BCELoss()
# hashing loss
contrast = Contrast_Loss(margin=args.margin)
quantization = Quantization_Loss()
best_value = 0
for epoch in range(1, args.EPOCH + 1):
G.train()
D.train()
step = 0
description = "training :" + str(epoch) + "/" + str(args.EPOCH)
with tqdm(trainloader, desc=description) as iterator:
for i, (data, target) in enumerate(iterator):
data, target = data.cuda(), target.cuda()
gap_softmax, softmax, hash_codes = G(data)
# ----------------------- calculate the total loss for hash learning ---------------------------
loss_images_softmax = crossentropy(softmax, target)
loss_gap_softmax = crossentropy(gap_softmax, target)
quan = quantization(hash_codes)
hinge = contrast(hash_codes, target)
hash_loss = hinge + args.alpha * loss_images_softmax + args.gamma * loss_gap_softmax + args.theta * quan
# ------------------------------- generate and adversarial stage -------------------------------
# ------------------------------- training the generate
g_loss = 0
if step % 5 == 0:
g_loss = adversarial_loss(D(hash_codes), valid)
loss = hash_loss + g_loss
optimizer_G.zero_grad()
loss.backward()
optimizer_G.step()
# ------------------------------- training the discriminator
if epoch < 2:
real_binary_data = generate_binary_distribution(
data.size(0), dim=args.Kbits)
real_binary = Variable(
torch.from_numpy(real_binary_data).type(
torch.FloatTensor),
requires_grad=False).cuda()
real_loss = adversarial_loss(D(real_binary), valid)
fake_loss = adversarial_loss(D(hash_codes.detach()), fake)
d_loss = real_loss + fake_loss
optimizer_D.zero_grad()
d_loss.backward()
optimizer_D.step()
# ------------------------ displaying the loss value during training ---------------------------
hash_softmax_correct = calculate_classification_accuracy(
softmax, target) / target.size(0)
cam_softmax_correct = calculate_classification_accuracy(
gap_softmax, target) / target.size(0)
information = "Loss: {:.4f}, Hash codes classification {:.2f}, cam classification {:.2f}".format(
loss.item(), hash_softmax_correct, cam_softmax_correct)
iterator.set_postfix_str(information)
if epoch % 10 == 0:
G.eval()
description = "testing"
accumulated_gap_classification, accumulated_hash_classification, accumulated_number = 0, 0, 0
with tqdm(testloader, desc=description) as iterator:
for i, (data, target) in enumerate(iterator):
data, target = data.cuda(), target.cuda()
gap_softmax, softmax, hash_codes = G(data)
# ------------------------ displaying the classification accuracy in testing data ---------------------------
accumulated_hash_classification += calculate_classification_accuracy(
softmax, target)
accumulated_gap_classification += calculate_classification_accuracy(
gap_softmax, target)
accumulated_number = accumulated_number + data.size(0)
information = "Testing, Hash codes classification {:.2f}, cam classification {:.2f}".format(
accumulated_hash_classification / accumulated_number,
accumulated_gap_classification / accumulated_number)
iterator.set_postfix_str(information)
# whether save the model parameters
if accumulated_hash_classification > best_value:
best_value = accumulated_hash_classification
if os.path.exists(args.parameters) is False:
os.makedirs(args.parameters)
print("saved parameters path is : " +
str(args.parameters))
torch.save(G.state_dict(), args.parameters + "/G.ckpt")
torch.save(D.state_dict(), args.parameters + "/D.ckpt")
print(
"**********************saved trained weights***************************"
)