-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdnn.py
128 lines (101 loc) · 5.02 KB
/
dnn.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
import numpy as np
import matplotlib.pyplot as plt
import torch
from PIL import Image
# load CIFAR-10 dataset
from torchvision import datasets, transforms
normalize = transforms.Normalize(mean=[0.507, 0.487, 0.441], std=[0.267, 0.256, 0.276])
transform = transforms.Compose([transforms.ToTensor(), normalize]) # normalize to (-1, 1)
trainset = datasets.CIFAR10('data', download=True, train=True, transform=transform)
testset = datasets.CIFAR10('data', download=True, train=False, transform=transform)
# define the network
# load resnet18
from torchvision import models
resnet18 = models.resnet18(weights=models.ResNet18_Weights.IMAGENET1K_V1)
resnet34 = models.resnet34(weights=models.ResNet34_Weights.IMAGENET1K_V1)
resnet50 = models.resnet50(weights=models.ResNet50_Weights.IMAGENET1K_V1)
vgg16 = models.vgg16(weights=models.VGG16_Weights.IMAGENET1K_V1)
mobileNet = models.mobilenet_v2(weights=models.MobileNet_V2_Weights.IMAGENET1K_V1)
# freeze the parameters
for param in resnet18.parameters():
param.requires_grad = False
for param in resnet34.parameters():
param.requires_grad = False
for param in resnet50.parameters():
param.requires_grad = False
for param in vgg16.parameters():
param.requires_grad = False
for param in mobileNet.parameters():
param.requires_grad = False
# remove the last average pooling layer
resnet18 = torch.nn.Sequential(*(list(resnet18.children())[:-1]))
resnet34 = torch.nn.Sequential(*(list(resnet34.children())[:-1]))
resnet50 = torch.nn.Sequential(*(list(resnet50.children())[:-1]))
vgg16 = torch.nn.Sequential(*(list(vgg16.children())[:-1]))
mobileNet = torch.nn.Sequential(*(list(mobileNet.children())[:-1]))
# Write all the layers in the network to file
with open('model_info.txt', 'w') as f:
f.write('ResNet18')
f.write(str(resnet18))
f.write('\n\n ResNet34')
f.write(str(resnet34))
f.write('\n\n ResNet50')
f.write(str(resnet50))
f.write('\n\n VGG16')
f.write(str(vgg16))
f.write('\n MobileNet')
f.write(str(mobileNet))
# test each of the models
from torch.utils.data import DataLoader
from tqdm import tqdm
from sklearn.metrics import accuracy_score
def test_model(model, trainset):
model.eval()
testloader = DataLoader(trainset, batch_size=128, shuffle=True)
output_total = []
labels_total = []
for images, labels in tqdm(testloader):
images, labels = images.cuda(), labels.cuda()
output = model(images)
output = output.cpu().detach().numpy()
output_total.append(output)
labels_total.append(labels.cpu().detach().numpy())
output_total = np.concatenate(output_total, axis=0)
labels = np.concatenate(labels_total, axis=0)
return output_total, labels
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--max_pool', type=int, default=0)
max_pool = parser.parse_args().max_pool
models = ['resnet18', 'resnet34', 'resnet50', 'vgg16', 'mobileNet']
for model in models:
net = globals()[model].cuda()
train_output, train_labels = test_model(net, trainset)
test_output, test_labels = test_model(net, testset)
# bilinear interpolation seems to be better used for dl model input, not output
# do max pooling to reduce the dimension
if max_pool == 1:
train_output = train_output.reshape(train_output.shape[0], -1)
test_output = test_output.reshape(test_output.shape[0], -1)
if model == 'resnet18' or model == 'resnet34': # (50000, 512, 1, 1)
train_output = torch.nn.functional.max_pool1d(torch.from_numpy(train_output), kernel_size=2)
test_output = torch.nn.functional.max_pool1d(torch.from_numpy(test_output), kernel_size=2)
elif model == 'resnet50': # (50000, 2048, 1, 1)
train_output = torch.nn.functional.max_pool1d(torch.from_numpy(train_output), kernel_size=8)
test_output = torch.nn.functional.max_pool1d(torch.from_numpy(test_output), kernel_size=8)
elif model == 'vgg16': # (50000, 512, 7, 7)
train_output = torch.nn.functional.max_pool1d(torch.from_numpy(train_output), kernel_size=98)
test_output = torch.nn.functional.max_pool1d(torch.from_numpy(test_output), kernel_size=98)
elif model == 'mobileNet': # (50000, 1280, 1, 1)
train_output = torch.nn.functional.max_pool1d(torch.from_numpy(train_output), kernel_size=5)
test_output = torch.nn.functional.max_pool1d(torch.from_numpy(test_output), kernel_size=5)
# convert it back to 50000 x 256 x 1 x 1
train_output = train_output.numpy()
test_output = test_output.numpy()
train_output = train_output.reshape(1020, 256, 1, 1)
test_output = test_output.reshape((1360-1020), 256, 1, 1)
np.save(f'train_{model}_data.npy', train_output)
np.save(f'train_{model}_labels.npy', train_labels)
np.save(f"test_{model}_data.npy", test_output)
np.save(f"test_{model}_labels.npy", test_labels)