in segmentation/model/cnsn_resnet.py [0:0]
def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
groups=1, width_per_group=64, replace_stride_with_dilation=None,
norm_layer=None, block_idxs=None, active_num=None, pos=None, beta=None, crop=None, cnsn_type=None, cn_pos=None):
super(ResNet, self).__init__()
if norm_layer is None:
norm_layer = nn.BatchNorm2d
self._norm_layer = norm_layer
self.inplanes = 64
self.dilation = 1
if replace_stride_with_dilation is None:
# each element in the tuple indicates if we should replace
# the 2x2 stride with a dilated convolution instead
replace_stride_with_dilation = [False, False, False]
if len(replace_stride_with_dilation) != 3:
raise ValueError("replace_stride_with_dilation should be None "
"or a 3-element tuple, got {}".format(replace_stride_with_dilation))
self.groups = groups
self.base_width = width_per_group
if block_idxs:
block_idxs = block_idxs.split('_')
block_idxs = list(map(int, block_idxs))
print('block_idxs: {}'.format(block_idxs))
if beta is not None:
print('beta: {}'.format(beta))
if crop is not None:
print('crop in 2 instance mode: {}'.format(crop))
self.block_idxs = block_idxs
if block_idxs and 0 in block_idxs:
self.img_cn = CrossNorm(crop=crop, beta=beta)
self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = norm_layer(self.inplanes)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
if block_idxs and 1 in block_idxs:
self.layer1 = self._make_layer(block, 64, layers[0], pos=pos, beta=beta, crop=crop, cnsn_type=cnsn_type, cn_pos=cn_pos, custom=True)
else:
self.layer1 = self._make_layer(block, 64, layers[0], custom=False)
if block_idxs and 2 in block_idxs:
self.layer2 = self._make_layer(block, 128, layers[1], stride=2, dilate=replace_stride_with_dilation[0], pos=pos, beta=beta, crop=crop, cnsn_type=cnsn_type, cn_pos=cn_pos, custom=True)
else:
self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
dilate=replace_stride_with_dilation[0], custom=False)
if block_idxs and 3 in block_idxs:
self.layer3 = self._make_layer(block, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1], pos=pos, beta=beta, crop=crop, cnsn_type=cnsn_type, cn_pos=cn_pos, custom=True)
else:
self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
dilate=replace_stride_with_dilation[1], custom=False)
if block_idxs and 4 in block_idxs:
self.layer4 = self._make_layer(block, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2], pos=pos, beta=beta, crop=crop, cnsn_type=cnsn_type, cn_pos=cn_pos, custom=True)
else:
self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
dilate=replace_stride_with_dilation[2], custom=False)
self.fc = nn.Linear(512 * block.expansion, num_classes)
self.cn_modules = []
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear) and m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, CrossNorm):
self.cn_modules.append(m)
if 'cn' in cnsn_type or cn_pos is not None:
self.cn_num = len(self.cn_modules)
assert self.cn_num > 0
print('cn_num: {}'.format(self.cn_num))
self.active_num = active_num
assert self.active_num > 0
print('active_num: {}'.format(self.active_num))
# Zero-initialize the last BN in each residual branch,
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
if zero_init_residual:
for m in self.modules():
if isinstance(m, Bottleneck):
nn.init.constant_(m.bn3.weight, 0)
elif isinstance(m, BasicBlock):
nn.init.constant_(m.bn2.weight, 0)