论文链接：https://arxiv.org/pdf/1904.01169.pdf 2020 IEEE TPAMI 2020的一篇文章
代码链接：https://github.com/Res2Net/Res2Net-PretrainedModels
主要思想：主要是为了增强resnet的multi-scale的能力，以往获得the multi-scale features的方法往往都是一种layer-wise manner，比如FPN，Res2Net constructe hierarchical residual-like connections within one single residual block，获得了multi-scale features使得resnet increases the range of receptive fields。
下面的结构图会非常清楚，
经过conv1*1之后的feature在channel上等分为四个部分，接下来的部分公式化如下
K表示3 × 3 convolution，最后将输出的yi concat起来，因为每经过一个K即conv3*3，就将获得一个更大的感受野，所以最后输出是a different number and different combination of receptive field sizes/scales，另外为了减少参数量和基于保留原始感受野的考虑，第一个x即x1后面并没有也接上conv3*3.
文中对scale dimension进行了分析，w可以理解为上面xi的channel数目，s可以理解为有多少个x。从图中可以看出增加s是可以带来收益的，这也可以res2net的multi-scale，获得一个long range receptive field.
文中在目标检测，实例分割等都进行了实验，证明res2net结构均可带来提升
res2net50_26w_4s结构：https://gist.github.com/breezelj/c694c127585e7586dc1952f6cc611849，较于resnet50
- 每个layer的downsample：x1是用AvgPool2d(kernel_size=3, stride=2, padding=1)进行的，xi(i != 1)是用figure2中那个conv3*3进行的（比如Conv2d(52, 52, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)), shortcut中的downsample通过Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)完成的
- 从代码中可以看出，在每个layer的第一个bottleneck没有采取文中的Ki(xi + yi−1),而是直接Ki(xi)，应该是因为要在这个bottleneck进行上一点说的downsample操作
代码(来自:https://github.com/Res2Net/Res2Net-PretrainedModels）：
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo
import torch
import torch.nn.functional as F
__all__ = ['Res2Net', 'res2net50']
model_urls = {
'res2net50_26w_4s': 'https://shanghuagao.oss-cn-beijing.aliyuncs.com/res2net/res2net50_26w_4s-06e79181.pth',
'res2net50_48w_2s': 'https://shanghuagao.oss-cn-beijing.aliyuncs.com/res2net/res2net50_48w_2s-afed724a.pth',
'res2net50_14w_8s': 'https://shanghuagao.oss-cn-beijing.aliyuncs.com/res2net/res2net50_14w_8s-6527dddc.pth',
'res2net50_26w_6s': 'https://shanghuagao.oss-cn-beijing.aliyuncs.com/res2net/res2net50_26w_6s-19041792.pth',
'res2net50_26w_8s': 'https://shanghuagao.oss-cn-beijing.aliyuncs.com/res2net/res2net50_26w_8s-2c7c9f12.pth',
'res2net101_26w_4s': 'https://shanghuagao.oss-cn-beijing.aliyuncs.com/res2net/res2net101_26w_4s-02a759a1.pth',
}
class Bottle2neck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, baseWidth=26, scale=4, stype='normal'):
""" Constructor
Args:
inplanes: input channel dimensionality
planes: output channel dimensionality
stride: conv stride. Replaces pooling layer.
downsample: None when stride = 1
baseWidth: basic width of conv3x3
scale: number of scale.
type: 'normal': normal set. 'stage': first block of a new stage.
"""
super(Bottle2neck, self).__init__()
width = int(math.floor(planes * (baseWidth / 64.0)))
self.conv1 = nn.Conv2d(inplanes, width * scale, kernel_size=1, bias=False)
self.bn1 = nn.BatchNorm2d(width * scale)
if scale == 1:
self.nums = 1
else:
self.nums = scale - 1
if stype == 'stage':
self.pool = nn.AvgPool2d(kernel_size=3, stride=stride, padding=1)
convs = []
bns = []
for i in range(self.nums):
convs.append(nn.Conv2d(width, width, kernel_size=3, stride=stride, padding=1, bias=False))
bns.append(nn.BatchNorm2d(width))
self.convs = nn.ModuleList(convs)
self.bns = nn.ModuleList(bns)
self.conv3 = nn.Conv2d(width * scale, planes * self.expansion, kernel_size=1, bias=False)
self.bn3 = nn.BatchNorm2d(planes * self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stype = stype
self.scale = scale
self.width = width
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
spx = torch.split(out, self.width, 1)
for i in range(self.nums):
if i == 0 or self.stype == 'stage': #stage表示每个layer的第一个Bottle2neck, normal表示其他的Bottle2neck
sp = spx[i] #可以看出在第一个Bottle2neck是没有采取文中的Ki(xi + yi−1),而是直接Ki(xi)
else:
sp = sp + spx[i]
sp = self.convs[i](sp)
sp = self.relu(self.bns[i](sp))
if i == 0:
out = sp
else:
out = torch.cat((out, sp), 1)
if self.scale != 1 and self.stype == 'normal':
out = torch.cat((out, spx[self.nums]), 1)
elif self.scale != 1 and self.stype == 'stage':
out = torch.cat((out, self.pool(spx[self.nums])), 1)
out = self.conv3(out)
out = self.bn3(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class Res2Net(nn.Module):
def __init__(self, block, layers, baseWidth=26, scale=4, num_classes=1000):
self.inplanes = 64
super(Res2Net, self).__init__()
self.baseWidth = baseWidth
self.scale = scale
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Linear(512 * block.expansion, num_classes)
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.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def _make_layer(self, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(planes * block.expansion),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample=downsample,
stype='stage', baseWidth=self.baseWidth, scale=self.scale))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes, baseWidth=self.baseWidth, scale=self.scale))
return nn.Sequential(*layers)
def forward(self, x):
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.avgpool(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
return x
def res2net50(pretrained=False, **kwargs):
"""Constructs a Res2Net-50 model.
Res2Net-50 refers to the Res2Net-50_26w_4s.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 6, 3], baseWidth=26, scale=4, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net50_26w_4s']))
return model
def res2net50_26w_4s(pretrained=False, **kwargs):
"""Constructs a Res2Net-50_26w_4s model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 6, 3], baseWidth=26, scale=4, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net50_26w_4s']))
return model
def res2net101_26w_4s(pretrained=False, **kwargs):
"""Constructs a Res2Net-50_26w_4s model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 23, 3], baseWidth=26, scale=4, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net101_26w_4s']))
return model
def res2net50_26w_6s(pretrained=False, **kwargs):
"""Constructs a Res2Net-50_26w_4s model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 6, 3], baseWidth=26, scale=6, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net50_26w_6s']))
return model
def res2net50_26w_8s(pretrained=False, **kwargs):
"""Constructs a Res2Net-50_26w_4s model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 6, 3], baseWidth=26, scale=8, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net50_26w_8s']))
return model
def res2net50_48w_2s(pretrained=False, **kwargs):
"""Constructs a Res2Net-50_48w_2s model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 6, 3], baseWidth=48, scale=2, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net50_48w_2s']))
return model
def res2net50_14w_8s(pretrained=False, **kwargs):
"""Constructs a Res2Net-50_14w_8s model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = Res2Net(Bottle2neck, [3, 4, 6, 3], baseWidth=14, scale=8, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['res2net50_14w_8s']))
return model
if __name__ == '__main__':
images = torch.rand(1, 3, 224, 224)
model = res2net50_26w_4s(pretrained=False)
out = model(images)
print(model(images).size())