import torch
import torch.nn as nn
import math
from .shakedrop import ShakeDrop
[docs]def conv3x3(in_planes, out_planes, stride=1):
"""
3x3 convolution with padding
"""
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)
[docs]class BasicBlock(nn.Module):
outchannel_ratio = 1
def __init__(self, inplanes, planes, stride=1, downsample=None, p_shakedrop=1.0):
super(BasicBlock, self).__init__()
self.bn1 = nn.BatchNorm2d(inplanes)
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn2 = nn.BatchNorm2d(planes)
self.conv2 = conv3x3(planes, planes)
self.bn3 = nn.BatchNorm2d(planes)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
self.shake_drop = ShakeDrop(p_shakedrop)
[docs] def forward(self, x):
out = self.bn1(x)
out = self.conv1(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn3(out)
out = self.shake_drop(out)
if self.downsample is not None:
shortcut = self.downsample(x)
featuremap_size = shortcut.size()[2:4]
else:
shortcut = x
featuremap_size = out.size()[2:4]
batch_size = out.size()[0]
residual_channel = out.size()[1]
shortcut_channel = shortcut.size()[1]
if residual_channel != shortcut_channel:
padding = torch.autograd.Variable(
torch.cuda.FloatTensor(batch_size, residual_channel - shortcut_channel, featuremap_size[0],
featuremap_size[1]).fill_(0))
out += torch.cat((shortcut, padding), dim=1)
else:
out += shortcut
return out
[docs]class Bottleneck(nn.Module):
outchannel_ratio = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, p_shakedrop=1.0):
super(Bottleneck, self).__init__()
self.bn1 = nn.BatchNorm2d(inplanes)
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv2 = nn.Conv2d(planes, (planes * 1), kernel_size=3, stride=stride,
padding=1, bias=False)
self.bn3 = nn.BatchNorm2d((planes * 1))
self.conv3 = nn.Conv2d((planes * 1), planes * Bottleneck.outchannel_ratio, kernel_size=1, bias=False)
self.bn4 = nn.BatchNorm2d(planes * Bottleneck.outchannel_ratio)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
self.shake_drop = ShakeDrop(p_shakedrop)
[docs] def forward(self, x):
out = self.bn1(x)
out = self.conv1(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn3(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn4(out)
out = self.shake_drop(out)
if self.downsample is not None:
shortcut = self.downsample(x)
featuremap_size = shortcut.size()[2:4]
else:
shortcut = x
featuremap_size = out.size()[2:4]
batch_size = out.size()[0]
residual_channel = out.size()[1]
shortcut_channel = shortcut.size()[1]
if residual_channel != shortcut_channel:
padding = torch.autograd.Variable(
torch.cuda.FloatTensor(batch_size, residual_channel - shortcut_channel, featuremap_size[0],
featuremap_size[1]).fill_(0))
out += torch.cat((shortcut, padding), dim=1)
else:
out += shortcut
return out
[docs]class PyramidNet(nn.Module):
def __init__(self, dataset, depth, alpha, n_classes, bottleneck=True):
super(PyramidNet, self).__init__()
self.dataset = dataset
if self.dataset.startswith('cifar'):
self.inplanes = 16
if bottleneck:
n = int((depth - 2) / 9)
block = Bottleneck
else:
n = int((depth - 2) / 6)
block = BasicBlock
self.addrate = alpha / (3 * n * 1.0)
self.ps_shakedrop = [1. - (1.0 - (0.5 / (3 * n)) * (i + 1)) for i in range(3 * n)]
self.input_featuremap_dim = self.inplanes
self.conv1 = nn.Conv2d(3, self.input_featuremap_dim, kernel_size=3, stride=1, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(self.input_featuremap_dim)
self.featuremap_dim = self.input_featuremap_dim
self.layer1 = self.pyramidal_make_layer(block, n)
self.layer2 = self.pyramidal_make_layer(block, n, stride=2)
self.layer3 = self.pyramidal_make_layer(block, n, stride=2)
self.final_featuremap_dim = self.input_featuremap_dim
self.bn_final = nn.BatchNorm2d(self.final_featuremap_dim)
self.relu_final = nn.ReLU(inplace=True)
self.avgpool = nn.AvgPool2d(8)
self.fc = nn.Linear(self.final_featuremap_dim, n_classes)
elif dataset == 'imagenet':
blocks = {18: BasicBlock, 34: BasicBlock, 50: Bottleneck, 101: Bottleneck, 152: Bottleneck, 200: Bottleneck}
layers = {18: [2, 2, 2, 2], 34: [3, 4, 6, 3], 50: [3, 4, 6, 3], 101: [3, 4, 23, 3], 152: [3, 8, 36, 3],
200: [3, 24, 36, 3]}
if layers.get(depth) is None:
if bottleneck == True:
blocks[depth] = Bottleneck
temp_cfg = int((depth - 2) / 12)
else:
blocks[depth] = BasicBlock
temp_cfg = int((depth - 2) / 8)
layers[depth] = [temp_cfg, temp_cfg, temp_cfg, temp_cfg]
#logger.info('=> the layer configuration for each stage is set to', layers[depth])
self.inplanes = 64
self.addrate = alpha / (sum(layers[depth]) * 1.0)
self.input_featuremap_dim = self.inplanes
self.conv1 = nn.Conv2d(3, self.input_featuremap_dim, kernel_size=7, stride=2, padding=3, bias=False)
self.bn1 = nn.BatchNorm2d(self.input_featuremap_dim)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.featuremap_dim = self.input_featuremap_dim
self.layer1 = self.pyramidal_make_layer(blocks[depth], layers[depth][0])
self.layer2 = self.pyramidal_make_layer(blocks[depth], layers[depth][1], stride=2)
self.layer3 = self.pyramidal_make_layer(blocks[depth], layers[depth][2], stride=2)
self.layer4 = self.pyramidal_make_layer(blocks[depth], layers[depth][3], stride=2)
self.final_featuremap_dim = self.input_featuremap_dim
self.bn_final = nn.BatchNorm2d(self.final_featuremap_dim)
self.relu_final = nn.ReLU(inplace=True)
self.avgpool = nn.AvgPool2d(7)
self.fc = nn.Linear(self.final_featuremap_dim, n_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
assert len(self.ps_shakedrop) == 0, self.ps_shakedrop
[docs] def pyramidal_make_layer(self, block, block_depth, stride=1):
downsample = None
if stride != 1: # or self.inplanes != int(round(featuremap_dim_1st)) * block.outchannel_ratio:
downsample = nn.AvgPool2d((2, 2), stride=(2, 2), ceil_mode=True)
layers = []
self.featuremap_dim = self.featuremap_dim + self.addrate
layers.append(block(self.input_featuremap_dim, int(round(self.featuremap_dim)), stride, downsample, p_shakedrop=self.ps_shakedrop.pop(0)))
for i in range(1, block_depth):
temp_featuremap_dim = self.featuremap_dim + self.addrate
layers.append(
block(int(round(self.featuremap_dim)) * block.outchannel_ratio, int(round(temp_featuremap_dim)), 1, p_shakedrop=self.ps_shakedrop.pop(0)))
self.featuremap_dim = temp_featuremap_dim
self.input_featuremap_dim = int(round(self.featuremap_dim)) * block.outchannel_ratio
return nn.Sequential(*layers)
[docs] def forward(self, x):
if self.dataset == 'cifar10' or self.dataset == 'cifar100':
x = self.conv1(x)
x = self.bn1(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.bn_final(x)
x = self.relu_final(x)
x = self.avgpool(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
elif self.dataset == 'imagenet':
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.bn_final(x)
x = self.relu_final(x)
x = self.avgpool(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
return x