upload
Browse files- ckpts/resnet/cifar10/Dense_SA_best.path.tar +3 -0
- ckpts/resnet/cifar10/FF/fisher_newcheckpoint.pth.tar +3 -0
- ckpts/resnet/cifar10/FF/fisher_neweval_result.pth.tar +3 -0
- ckpts/resnet/cifar10/FT/FTcheckpoint.pth.tar +3 -0
- ckpts/resnet/cifar10/FT/FTeval_result.pth.tar +3 -0
- ckpts/resnet/cifar10/GA/GAcheckpoint.pth.tar +3 -0
- ckpts/resnet/cifar10/GA/GAeval_result.pth.tar +3 -0
- ckpts/resnet/cifar10/IU/wfishercheckpoint.pth.tar +3 -0
- ckpts/resnet/cifar10/IU/wfishereval_result.pth.tar +3 -0
- ckpts/resnet/cifar10/l1_sparse/FT_prunecheckpoint.pth.tar +3 -0
- ckpts/resnet/cifar10/l1_sparse/FT_pruneeval_result.pth.tar +3 -0
- ckpts/resnet/cifar10/retrain/retraincheckpoint.pth.tar +3 -0
- ckpts/resnet/cifar10/retrain/retraineval_result.pth.tar +3 -0
- models/ResNet.py +460 -0
ckpts/resnet/cifar10/Dense_SA_best.path.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:7420c8d9158ebb948094e39a17cb22ec1bd269957a41930e7171b036baedd26c
|
3 |
+
size 89489613
|
ckpts/resnet/cifar10/FF/fisher_newcheckpoint.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fa0010529135002e07c81491bf1985212fa28470c6e77f0c90c1f6d659037731
|
3 |
+
size 44775689
|
ckpts/resnet/cifar10/FF/fisher_neweval_result.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5ce00ed2a06fbd1a20e8ece24df65f336af83b4dca24b7f305dff9437dc3d5ee
|
3 |
+
size 1199
|
ckpts/resnet/cifar10/FT/FTcheckpoint.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5bc1f381a0c2c0bdf35109b5b5195b90aa3826d20966df94f6cbf340f881f782
|
3 |
+
size 44775689
|
ckpts/resnet/cifar10/FT/FTeval_result.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:331b95208b29cc2202a4a92170cec0a8e3f52a338ee79c8f1437326056808c4f
|
3 |
+
size 1199
|
ckpts/resnet/cifar10/GA/GAcheckpoint.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:922ef6c25c17124489e6c0754c78c9d824179d34977d6206ac1387a75e7b5e6a
|
3 |
+
size 44775689
|
ckpts/resnet/cifar10/GA/GAeval_result.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:94266550c31fa9507cc09496cfc0f81f4017401827039a0b98150dbe7c249be1
|
3 |
+
size 1199
|
ckpts/resnet/cifar10/IU/wfishercheckpoint.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4aee7cbb80a5961ce44ea68a6534643835edf57c7518fe35faaabba8a33147df
|
3 |
+
size 44775689
|
ckpts/resnet/cifar10/IU/wfishereval_result.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:64b013e64a4db5a48085967e498b0e21f8d5ebd563804057a1fe3a1f11e9f597
|
3 |
+
size 1199
|
ckpts/resnet/cifar10/l1_sparse/FT_prunecheckpoint.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:55f9adfd0450ac743a8ab08558463d32ad32b624a21657cd0bae1e9f10cbc845
|
3 |
+
size 44775689
|
ckpts/resnet/cifar10/l1_sparse/FT_pruneeval_result.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0117231e0a39b0adc48cd5348a9de7a3deaadc1efbec87ae1aa076e1d6ff7721
|
3 |
+
size 1199
|
ckpts/resnet/cifar10/retrain/retraincheckpoint.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:41a05992d8ee665c9466f0468f3c447691b5e5cde015189e7e86b6f21359ca00
|
3 |
+
size 44775689
|
ckpts/resnet/cifar10/retrain/retraineval_result.pth.tar
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:db34432b11e493aee52501966d3e0bd0f4d4502e154231f463430bb930805912
|
3 |
+
size 1199
|
models/ResNet.py
ADDED
@@ -0,0 +1,460 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
|
4 |
+
# from torchvision.models.utils import load_state_dict_from_url
|
5 |
+
|
6 |
+
|
7 |
+
class NormalizeByChannelMeanStd(torch.nn.Module):
|
8 |
+
def __init__(self, mean, std):
|
9 |
+
super(NormalizeByChannelMeanStd, self).__init__()
|
10 |
+
if not isinstance(mean, torch.Tensor):
|
11 |
+
mean = torch.tensor(mean)
|
12 |
+
if not isinstance(std, torch.Tensor):
|
13 |
+
std = torch.tensor(std)
|
14 |
+
self.register_buffer("mean", mean)
|
15 |
+
self.register_buffer("std", std)
|
16 |
+
|
17 |
+
def forward(self, tensor):
|
18 |
+
return self.normalize_fn(tensor, self.mean, self.std)
|
19 |
+
|
20 |
+
def extra_repr(self):
|
21 |
+
return "mean={}, std={}".format(self.mean, self.std)
|
22 |
+
|
23 |
+
def normalize_fn(self, tensor, mean, std):
|
24 |
+
"""Differentiable version of torchvision.functional.normalize"""
|
25 |
+
# here we assume the color channel is in at dim=1
|
26 |
+
mean = mean[None, :, None, None]
|
27 |
+
std = std[None, :, None, None]
|
28 |
+
return tensor.sub(mean).div(std)
|
29 |
+
|
30 |
+
|
31 |
+
__all__ = [
|
32 |
+
"ResNet",
|
33 |
+
"resnet18",
|
34 |
+
"resnet34",
|
35 |
+
"resnet50",
|
36 |
+
"resnet101",
|
37 |
+
"resnet152",
|
38 |
+
"resnext50_32x4d",
|
39 |
+
"resnext101_32x8d",
|
40 |
+
"wide_resnet50_2",
|
41 |
+
"wide_resnet101_2",
|
42 |
+
]
|
43 |
+
|
44 |
+
|
45 |
+
model_urls = {
|
46 |
+
"resnet18": "https://download.pytorch.org/models/resnet18-5c106cde.pth",
|
47 |
+
"resnet34": "https://download.pytorch.org/models/resnet34-333f7ec4.pth",
|
48 |
+
"resnet50": "https://download.pytorch.org/models/resnet50-19c8e357.pth",
|
49 |
+
"resnet101": "https://download.pytorch.org/models/resnet101-5d3b4d8f.pth",
|
50 |
+
"resnet152": "https://download.pytorch.org/models/resnet152-b121ed2d.pth",
|
51 |
+
"resnext50_32x4d": "https://download.pytorch.org/models/resnext50_32x4d-7cdf4587.pth",
|
52 |
+
"resnext101_32x8d": "https://download.pytorch.org/models/resnext101_32x8d-8ba56ff5.pth",
|
53 |
+
"wide_resnet50_2": "https://download.pytorch.org/models/wide_resnet50_2-95faca4d.pth",
|
54 |
+
"wide_resnet101_2": "https://download.pytorch.org/models/wide_resnet101_2-32ee1156.pth",
|
55 |
+
}
|
56 |
+
|
57 |
+
|
58 |
+
def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
|
59 |
+
"""3x3 convolution with padding"""
|
60 |
+
return nn.Conv2d(
|
61 |
+
in_planes,
|
62 |
+
out_planes,
|
63 |
+
kernel_size=3,
|
64 |
+
stride=stride,
|
65 |
+
padding=dilation,
|
66 |
+
groups=groups,
|
67 |
+
bias=False,
|
68 |
+
dilation=dilation,
|
69 |
+
)
|
70 |
+
|
71 |
+
|
72 |
+
def conv1x1(in_planes, out_planes, stride=1):
|
73 |
+
"""1x1 convolution"""
|
74 |
+
return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
|
75 |
+
|
76 |
+
|
77 |
+
class BasicBlock(nn.Module):
|
78 |
+
expansion = 1
|
79 |
+
__constants__ = ["downsample"]
|
80 |
+
|
81 |
+
def __init__(
|
82 |
+
self,
|
83 |
+
inplanes,
|
84 |
+
planes,
|
85 |
+
stride=1,
|
86 |
+
downsample=None,
|
87 |
+
groups=1,
|
88 |
+
base_width=64,
|
89 |
+
dilation=1,
|
90 |
+
norm_layer=None,
|
91 |
+
):
|
92 |
+
super(BasicBlock, self).__init__()
|
93 |
+
if norm_layer is None:
|
94 |
+
norm_layer = nn.BatchNorm2d
|
95 |
+
if groups != 1 or base_width != 64:
|
96 |
+
raise ValueError("BasicBlock only supports groups=1 and base_width=64")
|
97 |
+
if dilation > 1:
|
98 |
+
raise NotImplementedError("Dilation > 1 not supported in BasicBlock")
|
99 |
+
# Both self.conv1 and self.downsample layers downsample the input when stride != 1
|
100 |
+
self.conv1 = conv3x3(inplanes, planes, stride)
|
101 |
+
self.bn1 = norm_layer(planes)
|
102 |
+
self.relu = nn.ReLU(inplace=True)
|
103 |
+
self.conv2 = conv3x3(planes, planes)
|
104 |
+
self.bn2 = norm_layer(planes)
|
105 |
+
self.downsample = downsample
|
106 |
+
self.stride = stride
|
107 |
+
|
108 |
+
def forward(self, x):
|
109 |
+
identity = x
|
110 |
+
|
111 |
+
out = self.conv1(x)
|
112 |
+
out = self.bn1(out)
|
113 |
+
out = self.relu(out)
|
114 |
+
|
115 |
+
out = self.conv2(out)
|
116 |
+
out = self.bn2(out)
|
117 |
+
|
118 |
+
if self.downsample is not None:
|
119 |
+
identity = self.downsample(x)
|
120 |
+
|
121 |
+
out += identity
|
122 |
+
out = self.relu(out)
|
123 |
+
|
124 |
+
return out
|
125 |
+
|
126 |
+
|
127 |
+
class Bottleneck(nn.Module):
|
128 |
+
expansion = 4
|
129 |
+
__constants__ = ["downsample"]
|
130 |
+
|
131 |
+
def __init__(
|
132 |
+
self,
|
133 |
+
inplanes,
|
134 |
+
planes,
|
135 |
+
stride=1,
|
136 |
+
downsample=None,
|
137 |
+
groups=1,
|
138 |
+
base_width=64,
|
139 |
+
dilation=1,
|
140 |
+
norm_layer=None,
|
141 |
+
):
|
142 |
+
super(Bottleneck, self).__init__()
|
143 |
+
if norm_layer is None:
|
144 |
+
norm_layer = nn.BatchNorm2d
|
145 |
+
width = int(planes * (base_width / 64.0)) * groups
|
146 |
+
# Both self.conv2 and self.downsample layers downsample the input when stride != 1
|
147 |
+
self.conv1 = conv1x1(inplanes, width)
|
148 |
+
self.bn1 = norm_layer(width)
|
149 |
+
self.conv2 = conv3x3(width, width, stride, groups, dilation)
|
150 |
+
self.bn2 = norm_layer(width)
|
151 |
+
self.conv3 = conv1x1(width, planes * self.expansion)
|
152 |
+
self.bn3 = norm_layer(planes * self.expansion)
|
153 |
+
self.relu = nn.ReLU(inplace=True)
|
154 |
+
self.downsample = downsample
|
155 |
+
self.stride = stride
|
156 |
+
|
157 |
+
def forward(self, x):
|
158 |
+
identity = x
|
159 |
+
|
160 |
+
out = self.conv1(x)
|
161 |
+
out = self.bn1(out)
|
162 |
+
out = self.relu(out)
|
163 |
+
|
164 |
+
out = self.conv2(out)
|
165 |
+
out = self.bn2(out)
|
166 |
+
out = self.relu(out)
|
167 |
+
|
168 |
+
out = self.conv3(out)
|
169 |
+
out = self.bn3(out)
|
170 |
+
|
171 |
+
if self.downsample is not None:
|
172 |
+
identity = self.downsample(x)
|
173 |
+
|
174 |
+
out += identity
|
175 |
+
out = self.relu(out)
|
176 |
+
|
177 |
+
return out
|
178 |
+
|
179 |
+
|
180 |
+
class ResNet(nn.Module):
|
181 |
+
def __init__(
|
182 |
+
self,
|
183 |
+
block,
|
184 |
+
layers,
|
185 |
+
num_classes=1000,
|
186 |
+
zero_init_residual=False,
|
187 |
+
groups=1,
|
188 |
+
width_per_group=64,
|
189 |
+
replace_stride_with_dilation=None,
|
190 |
+
norm_layer=None,
|
191 |
+
imagenet=False,
|
192 |
+
):
|
193 |
+
super(ResNet, self).__init__()
|
194 |
+
if norm_layer is None:
|
195 |
+
norm_layer = nn.BatchNorm2d
|
196 |
+
self._norm_layer = norm_layer
|
197 |
+
|
198 |
+
self.inplanes = 64
|
199 |
+
self.dilation = 1
|
200 |
+
if replace_stride_with_dilation is None:
|
201 |
+
# each element in the tuple indicates if we should replace
|
202 |
+
# the 2x2 stride with a dilated convolution instead
|
203 |
+
replace_stride_with_dilation = [False, False, False]
|
204 |
+
if len(replace_stride_with_dilation) != 3:
|
205 |
+
raise ValueError(
|
206 |
+
"replace_stride_with_dilation should be None "
|
207 |
+
"or a 3-element tuple, got {}".format(replace_stride_with_dilation)
|
208 |
+
)
|
209 |
+
self.groups = groups
|
210 |
+
self.base_width = width_per_group
|
211 |
+
|
212 |
+
print("The normalize layer is contained in the network")
|
213 |
+
self.normalize = NormalizeByChannelMeanStd(
|
214 |
+
mean=[0.4914, 0.4822, 0.4465], std=[0.2470, 0.2435, 0.2616]
|
215 |
+
)
|
216 |
+
|
217 |
+
if not imagenet:
|
218 |
+
self.conv1 = nn.Conv2d(
|
219 |
+
3, self.inplanes, kernel_size=3, stride=1, padding=1, bias=False
|
220 |
+
)
|
221 |
+
self.bn1 = norm_layer(self.inplanes)
|
222 |
+
self.relu = nn.ReLU(inplace=True)
|
223 |
+
self.maxpool = nn.Identity()
|
224 |
+
else:
|
225 |
+
self.conv1 = nn.Conv2d(
|
226 |
+
3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False
|
227 |
+
)
|
228 |
+
self.bn1 = nn.BatchNorm2d(self.inplanes)
|
229 |
+
self.relu = nn.ReLU(inplace=True)
|
230 |
+
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
|
231 |
+
|
232 |
+
self.layer1 = self._make_layer(block, 64, layers[0])
|
233 |
+
self.layer2 = self._make_layer(
|
234 |
+
block, 128, layers[1], stride=2, dilate=replace_stride_with_dilation[0]
|
235 |
+
)
|
236 |
+
self.layer3 = self._make_layer(
|
237 |
+
block, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1]
|
238 |
+
)
|
239 |
+
self.layer4 = self._make_layer(
|
240 |
+
block, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2]
|
241 |
+
)
|
242 |
+
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
|
243 |
+
self.fc = nn.Linear(512 * block.expansion, num_classes)
|
244 |
+
|
245 |
+
for m in self.modules():
|
246 |
+
if isinstance(m, nn.Conv2d):
|
247 |
+
nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
|
248 |
+
elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
|
249 |
+
nn.init.constant_(m.weight, 1)
|
250 |
+
nn.init.constant_(m.bias, 0)
|
251 |
+
|
252 |
+
# Zero-initialize the last BN in each residual branch,
|
253 |
+
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
|
254 |
+
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
|
255 |
+
if zero_init_residual:
|
256 |
+
for m in self.modules():
|
257 |
+
if isinstance(m, Bottleneck):
|
258 |
+
nn.init.constant_(m.bn3.weight, 0)
|
259 |
+
elif isinstance(m, BasicBlock):
|
260 |
+
nn.init.constant_(m.bn2.weight, 0)
|
261 |
+
|
262 |
+
def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
|
263 |
+
norm_layer = self._norm_layer
|
264 |
+
downsample = None
|
265 |
+
previous_dilation = self.dilation
|
266 |
+
if dilate:
|
267 |
+
self.dilation *= stride
|
268 |
+
stride = 1
|
269 |
+
if stride != 1 or self.inplanes != planes * block.expansion:
|
270 |
+
downsample = nn.Sequential(
|
271 |
+
conv1x1(self.inplanes, planes * block.expansion, stride),
|
272 |
+
norm_layer(planes * block.expansion),
|
273 |
+
)
|
274 |
+
|
275 |
+
layers = []
|
276 |
+
layers.append(
|
277 |
+
block(
|
278 |
+
self.inplanes,
|
279 |
+
planes,
|
280 |
+
stride,
|
281 |
+
downsample,
|
282 |
+
self.groups,
|
283 |
+
self.base_width,
|
284 |
+
previous_dilation,
|
285 |
+
norm_layer,
|
286 |
+
)
|
287 |
+
)
|
288 |
+
self.inplanes = planes * block.expansion
|
289 |
+
for _ in range(1, blocks):
|
290 |
+
layers.append(
|
291 |
+
block(
|
292 |
+
self.inplanes,
|
293 |
+
planes,
|
294 |
+
groups=self.groups,
|
295 |
+
base_width=self.base_width,
|
296 |
+
dilation=self.dilation,
|
297 |
+
norm_layer=norm_layer,
|
298 |
+
)
|
299 |
+
)
|
300 |
+
|
301 |
+
return nn.Sequential(*layers)
|
302 |
+
|
303 |
+
def _forward_impl(self, x):
|
304 |
+
# See note [TorchScript super()]
|
305 |
+
x = self.normalize(x)
|
306 |
+
|
307 |
+
x = self.conv1(x)
|
308 |
+
x = self.bn1(x)
|
309 |
+
x = self.relu(x)
|
310 |
+
x = self.maxpool(x)
|
311 |
+
|
312 |
+
x = self.layer1(x)
|
313 |
+
x = self.layer2(x)
|
314 |
+
x = self.layer3(x)
|
315 |
+
x = self.layer4(x)
|
316 |
+
|
317 |
+
x = self.avgpool(x)
|
318 |
+
x = torch.flatten(x, 1)
|
319 |
+
# print(x.shape)
|
320 |
+
x = self.fc(x)
|
321 |
+
|
322 |
+
return x
|
323 |
+
|
324 |
+
def forward(self, x):
|
325 |
+
return self._forward_impl(x)
|
326 |
+
|
327 |
+
|
328 |
+
def _resnet(arch, block, layers, pretrained, progress, **kwargs):
|
329 |
+
model = ResNet(block, layers, **kwargs)
|
330 |
+
if pretrained:
|
331 |
+
state_dict = load_state_dict_from_url(model_urls[arch], progress=progress)
|
332 |
+
model.load_state_dict(state_dict)
|
333 |
+
return model
|
334 |
+
|
335 |
+
|
336 |
+
def resnet18(pretrained=False, progress=True, **kwargs):
|
337 |
+
r"""ResNet-18 model from
|
338 |
+
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
|
339 |
+
|
340 |
+
Args:
|
341 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
342 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
343 |
+
"""
|
344 |
+
return _resnet("resnet18", BasicBlock, [2, 2, 2, 2], pretrained, progress, **kwargs)
|
345 |
+
|
346 |
+
|
347 |
+
def resnet34(pretrained=False, progress=True, **kwargs):
|
348 |
+
r"""ResNet-34 model from
|
349 |
+
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
|
350 |
+
|
351 |
+
Args:
|
352 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
353 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
354 |
+
"""
|
355 |
+
return _resnet("resnet34", BasicBlock, [3, 4, 6, 3], pretrained, progress, **kwargs)
|
356 |
+
|
357 |
+
|
358 |
+
def resnet50(pretrained=False, progress=True, **kwargs):
|
359 |
+
r"""ResNet-50 model from
|
360 |
+
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
|
361 |
+
|
362 |
+
Args:
|
363 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
364 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
365 |
+
"""
|
366 |
+
return _resnet("resnet50", Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs)
|
367 |
+
|
368 |
+
|
369 |
+
def resnet101(pretrained=False, progress=True, **kwargs):
|
370 |
+
r"""ResNet-101 model from
|
371 |
+
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
|
372 |
+
|
373 |
+
Args:
|
374 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
375 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
376 |
+
"""
|
377 |
+
return _resnet(
|
378 |
+
"resnet101", Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs
|
379 |
+
)
|
380 |
+
|
381 |
+
|
382 |
+
def resnet152(pretrained=False, progress=True, **kwargs):
|
383 |
+
r"""ResNet-152 model from
|
384 |
+
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
|
385 |
+
|
386 |
+
Args:
|
387 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
388 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
389 |
+
"""
|
390 |
+
return _resnet(
|
391 |
+
"resnet152", Bottleneck, [3, 8, 36, 3], pretrained, progress, **kwargs
|
392 |
+
)
|
393 |
+
|
394 |
+
|
395 |
+
def resnext50_32x4d(pretrained=False, progress=True, **kwargs):
|
396 |
+
r"""ResNeXt-50 32x4d model from
|
397 |
+
`"Aggregated Residual Transformation for Deep Neural Networks" <https://arxiv.org/pdf/1611.05431.pdf>`_
|
398 |
+
|
399 |
+
Args:
|
400 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
401 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
402 |
+
"""
|
403 |
+
kwargs["groups"] = 32
|
404 |
+
kwargs["width_per_group"] = 4
|
405 |
+
return _resnet(
|
406 |
+
"resnext50_32x4d", Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs
|
407 |
+
)
|
408 |
+
|
409 |
+
|
410 |
+
def resnext101_32x8d(pretrained=False, progress=True, **kwargs):
|
411 |
+
r"""ResNeXt-101 32x8d model from
|
412 |
+
`"Aggregated Residual Transformation for Deep Neural Networks" <https://arxiv.org/pdf/1611.05431.pdf>`_
|
413 |
+
|
414 |
+
Args:
|
415 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
416 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
417 |
+
"""
|
418 |
+
kwargs["groups"] = 32
|
419 |
+
kwargs["width_per_group"] = 8
|
420 |
+
return _resnet(
|
421 |
+
"resnext101_32x8d", Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs
|
422 |
+
)
|
423 |
+
|
424 |
+
|
425 |
+
def wide_resnet50_2(pretrained=False, progress=True, **kwargs):
|
426 |
+
r"""Wide ResNet-50-2 model from
|
427 |
+
`"Wide Residual Networks" <https://arxiv.org/pdf/1605.07146.pdf>`_
|
428 |
+
|
429 |
+
The model is the same as ResNet except for the bottleneck number of channels
|
430 |
+
which is twice larger in every block. The number of channels in outer 1x1
|
431 |
+
convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048
|
432 |
+
channels, and in Wide ResNet-50-2 has 2048-1024-2048.
|
433 |
+
|
434 |
+
Args:
|
435 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
436 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
437 |
+
"""
|
438 |
+
kwargs["width_per_group"] = 64 * 2
|
439 |
+
return _resnet(
|
440 |
+
"wide_resnet50_2", Bottleneck, [3, 4, 6, 3], pretrained, progress, **kwargs
|
441 |
+
)
|
442 |
+
|
443 |
+
|
444 |
+
def wide_resnet101_2(pretrained=False, progress=True, **kwargs):
|
445 |
+
r"""Wide ResNet-101-2 model from
|
446 |
+
`"Wide Residual Networks" <https://arxiv.org/pdf/1605.07146.pdf>`_
|
447 |
+
|
448 |
+
The model is the same as ResNet except for the bottleneck number of channels
|
449 |
+
which is twice larger in every block. The number of channels in outer 1x1
|
450 |
+
convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048
|
451 |
+
channels, and in Wide ResNet-50-2 has 2048-1024-2048.
|
452 |
+
|
453 |
+
Args:
|
454 |
+
pretrained (bool): If True, returns a model pre-trained on ImageNet
|
455 |
+
progress (bool): If True, displays a progress bar of the download to stderr
|
456 |
+
"""
|
457 |
+
kwargs["width_per_group"] = 64 * 2
|
458 |
+
return _resnet(
|
459 |
+
"wide_resnet101_2", Bottleneck, [3, 4, 23, 3], pretrained, progress, **kwargs
|
460 |
+
)
|