1.官方代码可自行下载
下载链接:https://github.com/dbolya/yolacthttps://github.com/dbolya/yolact2.实现
2.1 参考Res2Net的构建
下载链接:https://github.com/Res2Net/Res2Net-PretrainedModelshttps://github.com/Res2Net/Res2Net-PretrainedModels2.2 借鉴Res2Net中的Bottle2neck以及Res2Net类结构
2.2.1 Bottle2neck类源码
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':
sp = spx[i]
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
2.2.2 Res2Net类源码
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
这里的构建方式就不多做介绍了,大家自行调试就能明白了!
2.3添加Yolact新的主干结构(在backbone.py内完成)
需要说明一下,看过Yolact论文或者源码的都应该知道,Yolact并不是直接使用残差结构的,它是将残差网络和特征金字塔网络融合,因此残差结构也不是顺序堆叠起来的,而是将4层大的layer的输出放到一个列表里,再和后面的特征金字塔融合,具体的代码可以参考Yolact源码里的backbone.py文件,里面的ResNetBacobone类写的很清楚。
2.3.1根据Yolact的ResNetBackbone构建方式,构建我们自己的Res2NetBackbone
上面的两段代码直接Copy到Backbone.py里肯定是不对的,需要按照官方的构建方式来构建。怎么修改也不多做解释(自己调试也能看明白),直接放代码:
先是在Backbone.py内添加一个Bottle2neck的类:
# Res2net_Bottle2neck
class Bottle2neck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, baseWidth=26, scale=4, stype='normal',dilation=1):
""" 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,dilation=dilation)
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=dilation, bias=False,dilation=dilation))
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,dilation=dilation)
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':
sp = spx[i]
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
再从后面添加一个Res2NetBackbone的类:
# Res2netBackbone
class Res2NetBackbone(nn.Module):
def __init__(self, layers, baseWidth=26, scale=4, block=Bottle2neck,atrous_layers = []):
self.inplanes = 64
super(Res2NetBackbone, self).__init__()
self.baseWidth = baseWidth
self.scale = scale
self.layers = nn.ModuleList()
self.channels = []
self.dilation = 1
self.atrous_layers = atrous_layers
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._make_layer(block, 64, layers[0])
self._make_layer(block, 128, layers[1], stride=2)
self._make_layer(block, 256, layers[2], stride=2)
self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d(1)
self.backbone_modules = [m for m in self.modules() if isinstance(m, nn.Conv2d)]
def _make_layer(self, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
if len(self.layers) in self.atrous_layers:
self.dilation += 1
stride = 1
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))
layer = nn.Sequential(*layers)
self.channels.append(planes * block.expansion)
self.layers.append(layer)
return layer
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
outs = []
for layer in self.layers:
x = layer(x)
outs.append(x)
return tuple(outs)
def init_backbone(self, path):
""" Initializes the backbone weights for training. """
state_dict = torch.load(path)
# Replace layer1 -> layers.0 etc.
keys = list(state_dict)
for key in keys:
if key.startswith('layer'):
idx = int(key[5])
new_key = 'layers.' + str(idx - 1) + key[6:]
state_dict[new_key] = state_dict.pop(key)
# Note: Using strict=False is berry scary. Triple check this.
self.load_state_dict(state_dict, strict=False)
def add_layer(self, conv_channels=1024, downsample=2, depth=1, block=Bottle2neck):
""" Add a downsample layer to the backbone as per what SSD does. """
self._make_layer(block, conv_channels // block.expansion, blocks=depth, stride=downsample)
到目前为止已经修改好backbone.py文件了,你还需要修改config.py文件,config.py文件路径再data文件夹内。
2.4修改Config(在./data/config.py内完成)
2.4.1 导入backbone.py内的Res2NetBackbone模块
from backbone import ResNetBackbone, VGGBackbone, ResNetBackboneGN, DarkNetBackbone,Res2NetBackbone
这个直接在后面加上Res2NetBackbone就行了
2.4.2 添加自己的Config
在适当位置添加上以下信息(适当位置自己看就可以了,这里就不多说了)
首先是res2net101_backbone和res2net50_backbone:
res2net101_backbone = backbone_base.copy({
'name': 'Res2Net101',
'path': 'res2net101_26w_4s-02a759a1.pth',
'type': Res2NetBackbone,
'args': ([3, 4, 23, 3],),
'transform': resnet_transform,
'selected_layers': list(range(2, 8)),
'pred_scales': [[1]]*6,
'pred_aspect_ratios': [ [[0.66685089, 1.7073535, 0.87508774, 1.16524493, 0.49059086]] ] * 6,
})
res2net50_backbone = resnet101_backbone.copy({
'name': 'Res2Net50',
'path': 'res2net50_26w_4s-06e79181.pth',
'type': Res2NetBackbone,
'args': ([3, 4, 6, 3],),
'transform': resnet_transform,
})
注意:需要下载预训练的权重文件 res2net101_26w_4s-02a759a1.pth 和 res2net50_26w_4s-06e79181.pth ,把他们放在weights文件夹里,最好是下载下来,不然没有的话你需要再改yolact.py内容,而且训练起来会很慢,我会把这两个文件放到最后。
然后是你自己的训练用的config,请看我的是什么样子,同样放到适当的位置(自己找找就可以,比较简单)。
yolact_res2net101_config = yolact_base_config.copy({
'name': 'yolact_res2net101',
'max_size': 512,
'backbone': res2net101_backbone.copy({
'selected_layers': list(range(1, 4)),
'pred_scales': [[int(x[0] / yolact_base_config.max_size * 512)] for x in yolact_base_config.backbone.pred_scales],
'pred_aspect_ratios': yolact_base_config.backbone.pred_aspect_ratios,
'use_pixel_scales': True,
'preapply_sqrt': False,
'use_square_anchors': True, # This is for backward compatability with a bug
}),
})
yolact_res2net50_config = yolact_base_config.copy({
'name': 'yolact_res2net50',
'max_size': 512,
'backbone': res2net50_backbone.copy({
'selected_layers': list(range(1, 4)),
'pred_scales': [[int(x[0] / yolact_base_config.max_size * 512)] for x in yolact_base_config.backbone.pred_scales],
'pred_aspect_ratios': yolact_base_config.backbone.pred_aspect_ratios,
'use_pixel_scales': True,
'preapply_sqrt': False,
'use_square_anchors': True, # This is for backward compatability with a bug
}),
})
注意:我训练的时候图片尺寸改成了512,然后需要把对应的'pred_scales'改掉,官方给出的是550的尺寸,这些其实都无所谓,看自己的情况就可以。
到目前为止要修改的地方都完成了,就可以通过设置train.py里的Parameters训练了,怎么设置就不讲了,网上一搜教程很详细。
3.预训练的权重文件(百度网盘链接,长期有效)
链接:https://pan.baidu.com/s/1CjYYcNEH6BgUiowd6Sk0wQ
提取码:haiu
--来自百度网盘超级会员V2的分享
记载我自己修改网络结构的同时,希望对你也有帮助,有缘再见!