LN1996 commited on
Commit
f533343
1 Parent(s): a511429

Upload 13 files

Browse files
custom_resnet.py ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # model.py file
2
+
3
+ import torch
4
+ import torch.nn as nn
5
+ import torch.nn.functional as F
6
+
7
+ class BasicBlock(nn.Module):
8
+
9
+ def __init__(self, in_channels, out_channels, stride=1):
10
+ super(BasicBlock, self).__init__()
11
+
12
+ self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3,
13
+ stride=stride, padding=1, bias=False)
14
+ self.bn1 = nn.BatchNorm2d(out_channels)
15
+
16
+ self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3,
17
+ stride=1, padding=1, bias=False)
18
+ self.bn2 = nn.BatchNorm2d(out_channels)
19
+
20
+ self.relu = nn.ReLU()
21
+
22
+ def forward(self, x):
23
+ x = self.relu(self.bn1(self.conv1(x)))
24
+ x = self.relu(self.bn2(self.conv2(x)))
25
+ return x
26
+
27
+ class ResNet(nn.Module):
28
+ def __init__(self, block, num_classes=10):
29
+ super(ResNet, self).__init__()
30
+
31
+ self.preparation = nn.Sequential(
32
+ nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False),
33
+ nn.BatchNorm2d(64),
34
+ nn.ReLU()
35
+ )
36
+
37
+ self.layer1 = nn.Sequential(
38
+ nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1, bias=False),
39
+ nn.MaxPool2d(2, 2),
40
+ nn.BatchNorm2d(128),
41
+ nn.ReLU()
42
+ )
43
+
44
+ self.residual1 = block(128, 128, 1)
45
+
46
+ self.layer2 = nn.Sequential(
47
+ nn.Conv2d(128, 256, kernel_size=3, stride=1, padding=1, bias=False),
48
+ nn.MaxPool2d(2, 2),
49
+ nn.BatchNorm2d(256),
50
+ nn.ReLU()
51
+ )
52
+
53
+ self.layer3 = nn.Sequential(
54
+ nn.Conv2d(256, 512, kernel_size=3, stride=1, padding=1, bias=False),
55
+ nn.MaxPool2d(2, 2),
56
+ nn.BatchNorm2d(512),
57
+ nn.ReLU()
58
+ )
59
+
60
+ self.residual3 = block(512, 512, 1)
61
+
62
+ self.maxpool2d = nn.MaxPool2d(4, 4)
63
+ self.fc = nn.Linear(512, num_classes)
64
+
65
+ def forward(self, x):
66
+
67
+ x = self.preparation(x)
68
+
69
+ x = self.layer1(x)
70
+ res1 = self.residual1(x)
71
+ x = x + res1
72
+
73
+ x = self.layer2(x)
74
+
75
+ x = self.layer3(x)
76
+ res3 = self.residual3(x)
77
+ x = x + res3
78
+
79
+ x = self.maxpool2d(x)
80
+ x = x.view(x.size(0), -1)
81
+
82
+ x = self.fc(x)
83
+
84
+ return x
85
+
86
+ def Custom_ResNet():
87
+ return ResNet(BasicBlock, num_classes=10)
examples/test_0.jpg ADDED
examples/test_1.jpg ADDED
examples/test_2.jpg ADDED
examples/test_3.jpg ADDED
examples/test_4.jpg ADDED
examples/test_5.jpg ADDED
examples/test_6.jpg ADDED
examples/test_7.jpg ADDED
examples/test_8.jpg ADDED
examples/test_9.jpg ADDED
results/custom_resnet_trained.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d0950f4cb97b78183165e325857b6c057eddd54bc82df71831879611de4f3b42
3
+ size 26326759
utils.py ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # utils file
2
+
3
+ import matplotlib.pyplot as plt
4
+ import torch
5
+ from torchsummary import summary
6
+ from torchvision import transforms
7
+ import torchvision
8
+ import numpy as np
9
+ import cv2
10
+
11
+ cv2.setNumThreads(0)
12
+ cv2.ocl.setUseOpenCL(False)
13
+
14
+ class Cifar10SearchDataset(torchvision.datasets.CIFAR10):
15
+
16
+ def __init__(self, root="./data", train=True, download=True, transform=None):
17
+ super().__init__(root=root, train=train, download=download, transform=transform)
18
+
19
+ def __getitem__(self, index):
20
+ image, label = self.data[index], self.targets[index]
21
+
22
+ if self.transform is not None:
23
+ transformed = self.transform(image=image)
24
+ image = transformed["image"]
25
+
26
+ return image, label
27
+
28
+ import albumentations as A
29
+ from albumentations.pytorch.transforms import ToTensorV2
30
+
31
+ import numpy as np
32
+ import cv2
33
+
34
+ cv2.setNumThreads(0)
35
+ cv2.ocl.setUseOpenCL(False)
36
+
37
+ def augmentation_custom_resnet(data, mu=(0.49139968, 0.48215827, 0.44653124), sigma=(0.24703233, 0.24348505, 0.26158768), pad=4):
38
+
39
+ if data == 'Train':
40
+ transform = A.Compose([A.PadIfNeeded(min_height=32+pad,
41
+ min_width=32+pad,
42
+ border_mode=cv2.BORDER_CONSTANT,
43
+ value=np.mean(mu)),
44
+ A.RandomCrop(32, 32),
45
+ A.HorizontalFlip(p=0.5),
46
+ A.Cutout(max_h_size=8, max_w_size=8),
47
+ A.Normalize(mean=mu, std=sigma),
48
+ ToTensorV2()])
49
+ else:
50
+ transform = A.Compose([A.Normalize(mean=mu, std=sigma),
51
+ ToTensorV2()])
52
+
53
+ return transform