|
a |
|
b/opengait/modeling/backbones/resnet.py |
|
|
1 |
from torch.nn import functional as F |
|
|
2 |
import torch.nn as nn |
|
|
3 |
from torchvision.models.resnet import BasicBlock, Bottleneck, ResNet |
|
|
4 |
from ..modules import BasicConv2d |
|
|
5 |
|
|
|
6 |
|
|
|
7 |
block_map = {'BasicBlock': BasicBlock, |
|
|
8 |
'Bottleneck': Bottleneck} |
|
|
9 |
|
|
|
10 |
|
|
|
11 |
class ResNet9(ResNet): |
|
|
12 |
def __init__(self, block, channels=[32, 64, 128, 256], in_channel=1, layers=[1, 2, 2, 1], strides=[1, 2, 2, 1], maxpool=True): |
|
|
13 |
if block in block_map.keys(): |
|
|
14 |
block = block_map[block] |
|
|
15 |
else: |
|
|
16 |
raise ValueError( |
|
|
17 |
"Error type for -block-Cfg-, supported: 'BasicBlock' or 'Bottleneck'.") |
|
|
18 |
self.maxpool_flag = maxpool |
|
|
19 |
super(ResNet9, self).__init__(block, layers) |
|
|
20 |
|
|
|
21 |
# Not used # |
|
|
22 |
self.fc = None |
|
|
23 |
############ |
|
|
24 |
self.inplanes = channels[0] |
|
|
25 |
self.bn1 = nn.BatchNorm2d(self.inplanes) |
|
|
26 |
|
|
|
27 |
self.conv1 = BasicConv2d(in_channel, self.inplanes, 3, 1, 1) |
|
|
28 |
|
|
|
29 |
self.layer1 = self._make_layer( |
|
|
30 |
block, channels[0], layers[0], stride=strides[0], dilate=False) |
|
|
31 |
|
|
|
32 |
self.layer2 = self._make_layer( |
|
|
33 |
block, channels[1], layers[1], stride=strides[1], dilate=False) |
|
|
34 |
self.layer3 = self._make_layer( |
|
|
35 |
block, channels[2], layers[2], stride=strides[2], dilate=False) |
|
|
36 |
self.layer4 = self._make_layer( |
|
|
37 |
block, channels[3], layers[3], stride=strides[3], dilate=False) |
|
|
38 |
|
|
|
39 |
def _make_layer(self, block, planes, blocks, stride=1, dilate=False): |
|
|
40 |
if blocks >= 1: |
|
|
41 |
layer = super()._make_layer(block, planes, blocks, stride=stride, dilate=dilate) |
|
|
42 |
else: |
|
|
43 |
def layer(x): return x |
|
|
44 |
return layer |
|
|
45 |
|
|
|
46 |
def forward(self, x): |
|
|
47 |
x = self.conv1(x) |
|
|
48 |
x = self.bn1(x) |
|
|
49 |
x = self.relu(x) |
|
|
50 |
if self.maxpool_flag: |
|
|
51 |
x = self.maxpool(x) |
|
|
52 |
|
|
|
53 |
x = self.layer1(x) |
|
|
54 |
x = self.layer2(x) |
|
|
55 |
x = self.layer3(x) |
|
|
56 |
x = self.layer4(x) |
|
|
57 |
return x |
|
|
58 |
|