maskrcnn-benchmark 代码详解之　resnet.py

１Resnet 结构

　　Resnet 一般分为５个卷积(conv)层，每一层为一个stage。其中每一个stage中由不同数量的相同的block(区块)构成，这些区块的个数就是block_count, 第一个stage跟其他几个stage结构完全不同，也可以看做是由单独的区块构成的，因此由区块不停堆叠构成的第二层到第５层(即stage2-stage5或conv2-conv5)，分别定义为index1-index4.就像搭积木一样，这四个层可有基本的区块搭成。下图为resnet的基本结构：

　　以下代码通过控制区块的多少，搭建出不同的Resnet(包括Resnet50等）：

# -----------------------------------------------------------------------------
# Standard ResNet models
# -----------------------------------------------------------------------------
# ResNet-50 (包括所有的阶段)
# ResNet 分为５个阶段，但是第一个阶段都相同，变化是从第二个阶段开始的，所以下面的index是从第二个阶段开始编号的。其中block_count为该阶段区块的个数
ResNet50StagesTo5 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, False), (2, 4, False), (3, 6, False), (4, 3, True))
)
# ResNet-50 up to stage 4 (excludes stage 5)
ResNet50StagesTo4 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, False), (2, 4, False), (3, 6, True))
)
# ResNet-101 (including all stages)
ResNet101StagesTo5 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, False), (2, 4, False), (3, 23, False), (4, 3, True))
)
# ResNet-101 up to stage 4 (excludes stage 5)
ResNet101StagesTo4 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, False), (2, 4, False), (3, 23, True))
)
# ResNet-50-FPN (including all stages)
ResNet50FPNStagesTo5 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, True), (2, 4, True), (3, 6, True), (4, 3, True))
)
# ResNet-101-FPN (including all stages)
ResNet101FPNStagesTo5 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, True), (2, 4, True), (3, 23, True), (4, 3, True))
)
# ResNet-152-FPN (including all stages)
ResNet152FPNStagesTo5 = tuple(StageSpec(index=i, block_count=c, return_features=r)for (i, c, r) in ((1, 3, True), (2, 8, True), (3, 36, True), (4, 3, True))
)

　根据以上的不同组合方案，maskrcnn benchmark可以搭建起不同的backbone:

def _make_stage(transformation_module,in_channels,bottleneck_channels,out_channels,block_count,num_groups,stride_in_1x1,first_stride,dilation=1,dcn_config={}
):blocks = []stride = first_stride# 根据不同的配置，构造不同的卷基层for _ in range(block_count):blocks.append(transformation_module(in_channels,bottleneck_channels,out_channels,num_groups,stride_in_1x1,stride,dilation=dilation,dcn_config=dcn_config))stride = 1in_channels = out_channelsreturn nn.Sequential(*blocks)

这几种不同的backbone之后被集成为一个统一的对象以便于调用，其代码为:

_STAGE_SPECS = Registry({"R-50-C4": ResNet50StagesTo4,"R-50-C5": ResNet50StagesTo5,"R-101-C4": ResNet101StagesTo4,"R-101-C5": ResNet101StagesTo5,"R-50-FPN": ResNet50FPNStagesTo5,"R-50-FPN-RETINANET": ResNet50FPNStagesTo5,"R-101-FPN": ResNet101FPNStagesTo5,"R-101-FPN-RETINANET": ResNet101FPNStagesTo5,"R-152-FPN": ResNet152FPNStagesTo5,
})

２区块(block）结构

　　2.1 Bottleneck结构

　　刚刚提到，在Resnet中，第一层卷基层可以看做一种区块，而第二层到第五层由不同的称之为Bottleneck的区块堆叠二层。第一层可以看做一个stem区块。其中Bottleneck的结构如下：

　　在maskrcnn benchmark中构造以上结构的代码为：

class Bottleneck(nn.Module):def __init__(self,in_channels,bottleneck_channels,out_channels,num_groups,stride_in_1x1,stride,dilation,norm_func,dcn_config):super(Bottleneck, self).__init__()# 区块旁边的旁支self.downsample = Noneif in_channels != out_channels:# 获得卷积的步长　使用一个长度为１的卷积核对输入特征进行卷积，使得其输出通道数等于主体部分的输出通道数down_stride = stride if dilation == 1 else 1self.downsample = nn.Sequential(Conv2d(in_channels, out_channels,kernel_size=1, stride=down_stride, bias=False),norm_func(out_channels),)for modules in [self.downsample,]:for l in modules.modules():if isinstance(l, Conv2d):nn.init.kaiming_uniform_(l.weight, a=1)if dilation > 1:stride = 1 # reset to be 1# The original MSRA ResNet models have stride in the first 1x1 conv# The subsequent fb.torch.resnet and Caffe2 ResNe[X]t implementations have# stride in the 3x3 conv# 步长stride_1x1, stride_3x3 = (stride, 1) if stride_in_1x1 else (1, stride)#　区块中主体部分，这一部分为固定结构# 使得特征经过长度大小为１的卷积核self.conv1 = Conv2d(in_channels,bottleneck_channels,kernel_size=1,stride=stride_1x1,bias=False,)self.bn1 = norm_func(bottleneck_channels)# TODO: specify init for the abovewith_dcn = dcn_config.get("stage_with_dcn", False)if with_dcn:# 使用dcn网络deformable_groups = dcn_config.get("deformable_groups", 1)with_modulated_dcn = dcn_config.get("with_modulated_dcn", False)self.conv2 = DFConv2d(bottleneck_channels, bottleneck_channels, defrost=with_modulated_dcn,kernel_size=3, stride=stride_3x3, groups=num_groups,dilation=dilation,deformable_groups=deformable_groups,bias=False)else:# 使得特征经过长度大小为３的卷积核self.conv2 = Conv2d(bottleneck_channels,bottleneck_channels,kernel_size=3,stride=stride_3x3,padding=dilation,bias=False,groups=num_groups,dilation=dilation)nn.init.kaiming_uniform_(self.conv2.weight, a=1)self.bn2 = norm_func(bottleneck_channels)self.conv3 = Conv2d(bottleneck_channels, out_channels, kernel_size=1, bias=False)self.bn3 = norm_func(out_channels)for l in [self.conv1, self.conv3,]:nn.init.kaiming_uniform_(l.weight, a=1)def forward(self, x):identity = xout = self.conv1(x)out = self.bn1(out)out = F.relu_(out)out = self.conv2(out)out = self.bn2(out)out = F.relu_(out)out0 = self.conv3(out)out = self.bn3(out0)if self.downsample is not None:identity = self.downsample(x)out += identityout = F.relu_(out)return out

　　2.２ Stem结构

　　刚刚提到Resnet的第一层可以看做是一个Stem结构，其结构的代码为：

class BaseStem(nn.Module):def __init__(self, cfg, norm_func):super(BaseStem, self).__init__()# 获取backbone的输出特征层的输出通道数，由用户自定义out_channels = cfg.MODEL.RESNETS.STEM_OUT_CHANNELS# 输入通道数为图像的三原色，输出为输出通道数，这一部分是固定的，又Resnet论文定义的self.conv1 = Conv2d(3, out_channels, kernel_size=7, stride=2, padding=3, bias=False)self.bn1 = norm_func(out_channels)for l in [self.conv1,]:nn.init.kaiming_uniform_(l.weight, a=1)def forward(self, x):x = self.conv1(x)x = self.bn1(x)x = F.relu_(x)x = F.max_pool2d(x, kernel_size=3, stride=2, padding=1)return x

　　2.３两种结构的衍生与封装

　　在maskrcnn benchmark中，对上面提到的这两种block结构进行的衍生和封装，Bottleneck和Stem分别衍生出带有Batch Normalization 和 Group Normalizetion的封装类，分别为：BottleneckWithFixedBatchNorm, StemWithFixedBatchNorm, BottleneckWithGN, StemWithGN. 其代码过于简单，就不做注释：

class BottleneckWithFixedBatchNorm(Bottleneck):def __init__(self,in_channels,bottleneck_channels,out_channels,num_groups=1,stride_in_1x1=True,stride=1,dilation=1,dcn_config={}):super(BottleneckWithFixedBatchNorm, self).__init__(in_channels=in_channels,bottleneck_channels=bottleneck_channels,out_channels=out_channels,num_groups=num_groups,stride_in_1x1=stride_in_1x1,stride=stride,dilation=dilation,norm_func=FrozenBatchNorm2d,dcn_config=dcn_config)class StemWithFixedBatchNorm(BaseStem):def __init__(self, cfg):super(StemWithFixedBatchNorm, self).__init__(cfg, norm_func=FrozenBatchNorm2d)class BottleneckWithGN(Bottleneck):def __init__(self,in_channels,bottleneck_channels,out_channels,num_groups=1,stride_in_1x1=True,stride=1,dilation=1,dcn_config={}):super(BottleneckWithGN, self).__init__(in_channels=in_channels,bottleneck_channels=bottleneck_channels,out_channels=out_channels,num_groups=num_groups,stride_in_1x1=stride_in_1x1,stride=stride,dilation=dilation,norm_func=group_norm,dcn_config=dcn_config)class StemWithGN(BaseStem):def __init__(self, cfg):super(StemWithGN, self).__init__(cfg, norm_func=group_norm)_TRANSFORMATION_MODULES = Registry({"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm,"BottleneckWithGN": BottleneckWithGN,
})

　　接着，这两种结构关于BN和GN的四种衍生类被封装起来，以便于调用。其封装为：

_TRANSFORMATION_MODULES = Registry({"BottleneckWithFixedBatchNorm": BottleneckWithFixedBatchNorm,"BottleneckWithGN": BottleneckWithGN,
})_STEM_MODULES = Registry({"StemWithFixedBatchNorm": StemWithFixedBatchNorm,"StemWithGN": StemWithGN,
})

3 Resnet总体结构

　　３.1 Resnet结构

　　在以上的基础上，我们可以在以上结构上进一步搭建起真正的Resnet. 其中包括第一层卷基层，和其他四个阶段，代码为：

class ResNet(nn.Module):def __init__(self, cfg):super(ResNet, self).__init__()# If we want to use the cfg in forward(), then we should make a copy# of it and store it for later use:# self.cfg = cfg.clone()# Translate string names to implementations# 第一层conv层，也是第一阶段，以stem的形式展现stem_module = _STEM_MODULES[cfg.MODEL.RESNETS.STEM_FUNC]# 得到指定的backbone结构stage_specs = _STAGE_SPECS[cfg.MODEL.BACKBONE.CONV_BODY]# 　得到具体bottleneck结构，也就是指出组成backbone基本模块的类型transformation_module = _TRANSFORMATION_MODULES[cfg.MODEL.RESNETS.TRANS_FUNC]# Construct the stem moduleself.stem = stem_module(cfg)# Constuct the specified ResNet stages# 用于group normalization设置的组数num_groups = cfg.MODEL.RESNETS.NUM_GROUPS# 指定每一组拥有的通道数width_per_group = cfg.MODEL.RESNETS.WIDTH_PER_GROUP# stem是第一层的结构，它的输出也就是第二层一下的组合结构的输入通道数，内部通道数是可以自由定义的in_channels = cfg.MODEL.RESNETS.STEM_OUT_CHANNELS# 使用group的数目和每一组的通道数来得出组成backbone基本模块的内部通道数stage2_bottleneck_channels = num_groups * width_per_group# 第二阶段的输出通道数stage2_out_channels = cfg.MODEL.RESNETS.RES2_OUT_CHANNELSself.stages = []self.return_features = {}for stage_spec in stage_specs:name = "layer" + str(stage_spec.index)# 以下每一阶段的输入输出层的通道数都可以由stage2层的得到，即２倍关系stage2_relative_factor = 2 ** (stage_spec.index - 1)bottleneck_channels = stage2_bottleneck_channels * stage2_relative_factorout_channels = stage2_out_channels * stage2_relative_factorstage_with_dcn = cfg.MODEL.RESNETS.STAGE_WITH_DCN[stage_spec.index -1]# 得到每一阶段的卷积结构module = _make_stage(transformation_module,in_channels,bottleneck_channels,out_channels,stage_spec.block_count,num_groups,cfg.MODEL.RESNETS.STRIDE_IN_1X1,first_stride=int(stage_spec.index > 1) + 1,dcn_config={"stage_with_dcn": stage_with_dcn,"with_modulated_dcn": cfg.MODEL.RESNETS.WITH_MODULATED_DCN,"deformable_groups": cfg.MODEL.RESNETS.DEFORMABLE_GROUPS,})in_channels = out_channelsself.add_module(name, module)self.stages.append(name)self.return_features[name] = stage_spec.return_features# Optionally freeze (requires_grad=False) parts of the backboneself._freeze_backbone(cfg.MODEL.BACKBONE.FREEZE_CONV_BODY_AT)# 　   固定某一层的参数不再更新def _freeze_backbone(self, freeze_at):if freeze_at < 0:returnfor stage_index in range(freeze_at):if stage_index == 0:m = self.stem  # stage 0 is the stemelse:m = getattr(self, "layer" + str(stage_index))for p in m.parameters():p.requires_grad = Falsedef forward(self, x):outputs = []x = self.stem(x)for stage_name in self.stages:x = getattr(self, stage_name)(x)if self.return_features[stage_name]:outputs.append(x)return outputs

　　３.２ Resnet head结构

Head，在我理解看来就是完成某种功能的网络结构，Resnet head就是指使用Bottleneck块堆叠成不同的用于构成Resnet的功能网络结构，它内部结构相似，完成某种功能。在此不做过多介绍，因为是上面的Resnet子结构

class ResNetHead(nn.Module):def __init__(self,block_module,stages,num_groups=1,width_per_group=64,stride_in_1x1=True,stride_init=None,res2_out_channels=256,dilation=1,dcn_config={}):super(ResNetHead, self).__init__()stage2_relative_factor = 2 ** (stages[0].index - 1)stage2_bottleneck_channels = num_groups * width_per_groupout_channels = res2_out_channels * stage2_relative_factorin_channels = out_channels // 2bottleneck_channels = stage2_bottleneck_channels * stage2_relative_factorblock_module = _TRANSFORMATION_MODULES[block_module]self.stages = []stride = stride_initfor stage in stages:name = "layer" + str(stage.index)if not stride:stride = int(stage.index > 1) + 1module = _make_stage(block_module,in_channels,bottleneck_channels,out_channels,stage.block_count,num_groups,stride_in_1x1,first_stride=stride,dilation=dilation,dcn_config=dcn_config)stride = Noneself.add_module(name, module)self.stages.append(name)self.out_channels = out_channelsdef forward(self, x):for stage in self.stages:x = getattr(self, stage)(x)return x