RepVGG：算法简介及repvgg.py代码解析

论文链接：RepVGG: Making VGG-style ConvNets Great Again

代码链接：https://github.com/DingXiaoH/RepVGG

一、论文亮点

在ResNet提出之前，计算机视觉模型大多是以VGG为代表的单分支模型，ResNet及其变种的多分支结构，一定程度上解决了梯度消失、模型退化等问题，但是，多分支结构对于训练阶段十分有益，但是在推理/部署时我们希望模型能更快、更灵活、更省内存。于是，这篇论文提出了一种称为“结构重参数化”(Structural Re-parameterization)，它在推理阶段时将训练阶段的模型进行等效融合，使其变成单分支结构，以便更加适应推理场景的需求。

对于论文的算法具体思路，可以参考一下这篇博文：RepVGG网络简介，本文侧重于逐句解析官方源码内容。

二、官方源码解析(repvgg.py)

import torch.nn as nn
import numpy as np
import torch
import copy
from se_block import SEBlockdef conv_bn(in_channels, out_channels, kernel_size, stride, padding, groups=1):"""带BN的卷积层"""result = nn.Sequential()result.add_module('conv', nn.Conv2d(in_channels=in_channels, out_channels=out_channels,kernel_size=kernel_size, stride=stride, padding=padding, groups=groups, bias=False))result.add_module('bn', nn.BatchNorm2d(num_features=out_channels))return resultclass RepVGGBlock(nn.Module):def __init__(self, in_channels, out_channels, kernel_size,stride=1, padding=0, dilation=1, groups=1, padding_mode='zeros', deploy=False, use_se=False):super(RepVGGBlock, self).__init__()self.deploy = deploy  # 推理部署self.groups = groupsself.in_channels = in_channelsassert kernel_size == 3assert padding == 1padding_11 = padding - kernel_size // 2  # //为对商进行向下取整；padding_11应该是1×1卷积的padding数self.nonlinearity = nn.ReLU()if use_se:  # 是否将identity分支换成SEself.se = SEBlock(out_channels, internal_neurons=out_channels // 16)else:self.se = nn.Identity()if deploy:  # 定义推理模型时，基本block就是一个简单的 conv2dself.rbr_reparam = nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride,padding=padding, dilation=dilation, groups=groups, bias=True, padding_mode=padding_mode)else:  # 训练时# identity分支，仅有一个BN层；当输入输出channel不同时，不采用identity分支，即只有3×3卷积分支和1×1卷积分支self.rbr_identity = nn.BatchNorm2d(num_features=in_channels) if out_channels == in_channels and stride == 1 else None# 3×3卷积层+BNself.rbr_dense = conv_bn(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride, padding=padding, groups=groups)# 1×1卷积+BN，这里padding_11应该为0self.rbr_1x1 = conv_bn(in_channels=in_channels, out_channels=out_channels, kernel_size=1, stride=stride, padding=padding_11, groups=groups)print('RepVGG Block, identity = ', self.rbr_identity)def forward(self, inputs):if hasattr(self, 'rbr_reparam'):  # 如果在推理阶段，会定义self.rbr_reparam这个属性，这里是判断是否在推理阶段# 注意：执行return语句后就会退出函数，之后的语句不再执行return self.nonlinearity(self.se(self.rbr_reparam(inputs)))  # self.se()要看use_se是否为True，决定是否用SE模块if self.rbr_identity is None:id_out = 0else:id_out = self.rbr_identity(inputs)# 训练阶段：3×3卷积、1×1卷积、identity三个分支的结果相加后进行ReLUreturn self.nonlinearity(self.se(self.rbr_dense(inputs) + self.rbr_1x1(inputs) + id_out))#   Optional. This improves the accuracy and facilitates quantization.#   1.  Cancel the original weight decay on rbr_dense.conv.weight and rbr_1x1.conv.weight.#   2.  Use like this.#       loss = criterion(....)#       for every RepVGGBlock blk:#           loss += weight_decay_coefficient * 0.5 * blk.get_cust_L2()#       optimizer.zero_grad()#       loss.backward()def get_custom_L2(self):K3 = self.rbr_dense.conv.weightK1 = self.rbr_1x1.conv.weightt3 = (self.rbr_dense.bn.weight / ((self.rbr_dense.bn.running_var + self.rbr_dense.bn.eps).sqrt())).reshape(-1, 1, 1, 1).detach()t1 = (self.rbr_1x1.bn.weight / ((self.rbr_1x1.bn.running_var + self.rbr_1x1.bn.eps).sqrt())).reshape(-1, 1, 1, 1).detach()l2_loss_circle = (K3 ** 2).sum() - (K3[:, :, 1:2, 1:2] ** 2).sum()      # The L2 loss of the "circle" of weights in 3x3 kernel. Use regular L2 on them.eq_kernel = K3[:, :, 1:2, 1:2] * t3 + K1 * t1                           # The equivalent resultant central point of 3x3 kernel.l2_loss_eq_kernel = (eq_kernel ** 2 / (t3 ** 2 + t1 ** 2)).sum()        # Normalize for an L2 coefficient comparable to regular L2.return l2_loss_eq_kernel + l2_loss_circle#   This func derives the equivalent kernel and bias in a DIFFERENTIABLE way.
#   You can get the equivalent kernel and bias at any time and do whatever you want,#   for example, apply some penalties or constraints during training, just like you do to the other models.
#   May be useful for quantization or pruning.def get_equivalent_kernel_bias(self):"""获取带BN的3×3卷积、带BN的1×1卷积和带BN的identity分支的等效卷积核、偏置，论文Fig4的第二个箭头"""kernel3x3, bias3x3 = self._fuse_bn_tensor(self.rbr_dense)kernel1x1, bias1x1 = self._fuse_bn_tensor(self.rbr_1x1)kernelid, biasid = self._fuse_bn_tensor(self.rbr_identity)# 返回三个等效卷积层的叠加，即三个卷积核相加、三个偏置相加return kernel3x3 + self._pad_1x1_to_3x3_tensor(kernel1x1) + kernelid, bias3x3 + bias1x1 + biasiddef _pad_1x1_to_3x3_tensor(self, kernel1x1):"""                     [0  0  0][1]  >>>padding>>>  [0  1  0][0  0  0] """if kernel1x1 is None:return 0else:return torch.nn.functional.pad(kernel1x1, [1,1,1,1])def _fuse_bn_tensor(self, branch):"""融合BN层，既可以实现卷积层和BN层融合成等效的3×3卷积层，也可以实现单独BN层等效成3×3卷积层，论文Fig4的第一个箭头"""if branch is None:  # 若branch不是3x3、1x1、BN，那就返回 W=0, b=0return 0, 0if isinstance(branch, nn.Sequential):  # 若branch是nn.Sequential()类型，即一个卷积层+BN层kernel = branch.conv.weight  # 卷积核权重running_mean = branch.bn.running_mean  # BN的均值running_var = branch.bn.running_var  # BN的方差gamma = branch.bn.weight  # BN的伽马beta = branch.bn.bias  # BN的偏置eps = branch.bn.eps  # BN的小参数，为了防止分母为0else:  # 如果传进来的仅仅是一个BN层assert isinstance(branch, nn.BatchNorm2d)if not hasattr(self, 'id_tensor'):input_dim = self.in_channels // self.groups  # self.group如果不等于1，即为分组卷积# 由于BN层不改变特征层维度，所以这里在自定义kernel时：batch=in_channels，那么输出特征层的深度依然是in_channels# np.zeros((self.in_channels, input_dim, 3, 3)即为in_channels个(input_dim, 3, 3)大小的卷积核# 每个卷积核生成一个channel的feature map，故生成输出feature map的通道数依然是in_channelskernel_value = np.zeros((self.in_channels, input_dim, 3, 3), dtype=np.float32)  # [batch,C,H,W]for i in range(self.in_channels):# 这个地方口头表达不太好说，总之就是将BN等效成3×3卷积核的过程，可以去看一下论文理解# 值得注意的是，i是从0算起的kernel_value[i, i % input_dim, 1, 1] = 1self.id_tensor = torch.from_numpy(kernel_value).to(branch.weight.device)  # 转成tensor并指定给device# 获取参数kernel = self.id_tensorrunning_mean = branch.running_meanrunning_var = branch.running_vargamma = branch.weightbeta = branch.biaseps = branch.epsstd = (running_var + eps).sqrt()  # 标准差t = (gamma / std).reshape(-1, 1, 1, 1)  # reshape成与[batch,1,1,1]方便在kernel * t时进行广播return kernel * t, beta - running_mean * gamma / std  # 返回加权后的kernel和偏置def switch_to_deploy(self):"""转换成推理所需的VGG结构"""if hasattr(self, 'rbr_reparam'):returnkernel, bias = self.get_equivalent_kernel_bias()self.rbr_reparam = nn.Conv2d(in_channels=self.rbr_dense.conv.in_channels, out_channels=self.rbr_dense.conv.out_channels,kernel_size=self.rbr_dense.conv.kernel_size, stride=self.rbr_dense.conv.stride,padding=self.rbr_dense.conv.padding, dilation=self.rbr_dense.conv.dilation, groups=self.rbr_dense.conv.groups, bias=True)self.rbr_reparam.weight.data = kernelself.rbr_reparam.bias.data = biasfor para in self.parameters():para.detach_()self.__delattr__('rbr_dense')self.__delattr__('rbr_1x1')if hasattr(self, 'rbr_identity'):self.__delattr__('rbr_identity')if hasattr(self, 'id_tensor'):self.__delattr__('id_tensor')self.deploy = Trueclass RepVGG(nn.Module):def __init__(self, num_blocks, num_classes=1000, width_multiplier=None, override_groups_map=None,deploy=False, use_se=False):super(RepVGG, self).__init__()# len(width_multiplier) == 4对应论文3.4节第三段第一句assert len(width_multiplier) == 4  # 宽度因子，改变输入/输出通道数self.deploy = deployself.override_groups_map = override_groups_map or dict()  # 用于分组卷积的字典，在L215处有定义self.use_se = use_seassert 0 not in self.override_groups_map  # 确保0不会成为分组数self.in_planes = min(64, int(64 * width_multiplier[0]))self.stage0 = RepVGGBlock(in_channels=3, out_channels=self.in_planes, kernel_size=3, stride=2, padding=1, deploy=self.deploy, use_se=self.use_se)self.cur_layer_idx = 1  # 当前layer的索引，用于对特定layer设置分组卷积self.stage1 = self._make_stage(int(64 * width_multiplier[0]), num_blocks[0], stride=2)self.stage2 = self._make_stage(int(128 * width_multiplier[1]), num_blocks[1], stride=2)self.stage3 = self._make_stage(int(256 * width_multiplier[2]), num_blocks[2], stride=2)self.stage4 = self._make_stage(int(512 * width_multiplier[3]), num_blocks[3], stride=2)self.gap = nn.AdaptiveAvgPool2d(output_size=1)self.linear = nn.Linear(int(512 * width_multiplier[3]), num_classes)def _make_stage(self, planes, num_blocks, stride):strides = [stride] + [1]*(num_blocks-1)  # 论文中提到：每个stage仅在第一个layer进行步长为2blocks = []  # 创建空列表for stride in strides:# 获取override_groups_map中键self.cur_layer_idx对应的值，若未设置值，则返回默认值1cur_groups = self.override_groups_map.get(self.cur_layer_idx, 1)  # 分组数blocks.append(RepVGGBlock(in_channels=self.in_planes, out_channels=planes, kernel_size=3,stride=stride, padding=1, groups=cur_groups, deploy=self.deploy, use_se=self.use_se))self.in_planes = planes  # RepVGG中实例属性self.in_planes随传入planes而改变self.cur_layer_idx += 1  # 每次循环即创建一个layer，那么相应的layer的索引就要+1return nn.Sequential(*blocks)def forward(self, x):  # 前向传播out = self.stage0(x)out = self.stage1(out)out = self.stage2(out)out = self.stage3(out)out = self.stage4(out)out = self.gap(out)out = out.view(out.size(0), -1)out = self.linear(out)return out# 特定layer的索引分组卷积
optional_groupwise_layers = [2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26]  # 论文3.4节第四段
g2_map = {l: 2 for l in optional_groupwise_layers}  # {2: 2, 4: 2, 6: 2,...,20: 2, 22: 2, 24: 2, 26: 2}
g4_map = {l: 4 for l in optional_groupwise_layers}def create_RepVGG_A0(deploy=False):return RepVGG(num_blocks=[2, 4, 14, 1], num_classes=1000,width_multiplier=[0.75, 0.75, 0.75, 2.5], override_groups_map=None, deploy=deploy)def create_RepVGG_A1(deploy=False):return RepVGG(num_blocks=[2, 4, 14, 1], num_classes=1000,width_multiplier=[1, 1, 1, 2.5], override_groups_map=None, deploy=deploy)def create_RepVGG_A2(deploy=False):return RepVGG(num_blocks=[2, 4, 14, 1], num_classes=1000,width_multiplier=[1.5, 1.5, 1.5, 2.75], override_groups_map=None, deploy=deploy)def create_RepVGG_B0(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[1, 1, 1, 2.5], override_groups_map=None, deploy=deploy)def create_RepVGG_B1(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[2, 2, 2, 4], override_groups_map=None, deploy=deploy)def create_RepVGG_B1g2(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[2, 2, 2, 4], override_groups_map=g2_map, deploy=deploy)def create_RepVGG_B1g4(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[2, 2, 2, 4], override_groups_map=g4_map, deploy=deploy)def create_RepVGG_B2(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=None, deploy=deploy)def create_RepVGG_B2g2(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=g2_map, deploy=deploy)def create_RepVGG_B2g4(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=g4_map, deploy=deploy)def create_RepVGG_B3(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[3, 3, 3, 5], override_groups_map=None, deploy=deploy)def create_RepVGG_B3g2(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[3, 3, 3, 5], override_groups_map=g2_map, deploy=deploy)def create_RepVGG_B3g4(deploy=False):return RepVGG(num_blocks=[4, 6, 16, 1], num_classes=1000,width_multiplier=[3, 3, 3, 5], override_groups_map=g4_map, deploy=deploy)def create_RepVGG_D2se(deploy=False):return RepVGG(num_blocks=[8, 14, 24, 1], num_classes=1000,width_multiplier=[2.5, 2.5, 2.5, 5], override_groups_map=None, deploy=deploy, use_se=True)func_dict = {
'RepVGG-A0': create_RepVGG_A0,
'RepVGG-A1': create_RepVGG_A1,
'RepVGG-A2': create_RepVGG_A2,
'RepVGG-B0': create_RepVGG_B0,
'RepVGG-B1': create_RepVGG_B1,
'RepVGG-B1g2': create_RepVGG_B1g2,
'RepVGG-B1g4': create_RepVGG_B1g4,
'RepVGG-B2': create_RepVGG_B2,
'RepVGG-B2g2': create_RepVGG_B2g2,
'RepVGG-B2g4': create_RepVGG_B2g4,
'RepVGG-B3': create_RepVGG_B3,
'RepVGG-B3g2': create_RepVGG_B3g2,
'RepVGG-B3g4': create_RepVGG_B3g4,
'RepVGG-D2se': create_RepVGG_D2se,      #   Updated at April 25, 2021. This is not reported in the CVPR paper.
}def get_RepVGG_func_by_name(name):  # 传入所需构建的model名（key），返回构建model的函数return func_dict[name]#   Use this for converting a RepVGG model or a bigger model with RepVGG as its component
#   Use like this
#   model = create_RepVGG_A0(deploy=False)
#   train model or load weights
#   repvgg_model_convert(model, save_path='repvgg_deploy.pth')
#   If you want to preserve the original model, call with do_copy=True#   ====================== for using RepVGG as the backbone of a bigger model, e.g., PSPNet, the pseudo code will be like
#   train_backbone = create_RepVGG_B2(deploy=False)
#   train_backbone.load_state_dict(torch.load('RepVGG-B2-train.pth'))
#   train_pspnet = build_pspnet(backbone=train_backbone)
#   segmentation_train(train_pspnet)
#   deploy_pspnet = repvgg_model_convert(train_pspnet)
#   segmentation_test(deploy_pspnet)
#   =====================   example_pspnet.py shows an exampledef repvgg_model_convert(model:torch.nn.Module, save_path=None, do_copy=True):"""模型转换"""if do_copy:model = copy.deepcopy(model)for module in model.modules():if hasattr(module, 'switch_to_deploy'):module.switch_to_deploy()if save_path is not None:torch.save(model.state_dict(), save_path)return model

------------------------------------废话分割线------------------------------------

浔阳江头夜送客，枫叶荻花秋瑟瑟。主人下马客在船，举酒欲饮无管弦。醉不成欢惨将别，别时茫茫江浸月。

忽闻水上琵琶声，主人忘归客不发。寻声暗问弹者谁，琵琶声停欲语迟。移船相近邀相见，添酒回灯重开宴。千呼万唤始出来，犹抱琵琶半遮面。转轴拨弦三两声，未成曲调先有情。弦弦掩抑声声思，似诉平生不得志。低眉信手续续弹，说尽心中无限事。轻拢慢捻抹复挑，初为《霓裳》后《六幺》。大弦嘈嘈如急雨，小弦切切如私语。嘈嘈切切错杂弹，大珠小珠落玉盘。间关莺语花底滑，幽咽泉流冰下难。冰泉冷涩弦凝绝，凝绝不通声暂歇。别有幽愁暗恨生，此时无声胜有声。银瓶乍破水浆迸，铁骑突出刀枪鸣。曲终收拨当心画，四弦一声如裂帛。东船西舫悄无言，唯见江心秋月白。

沉吟放拨插弦中，整顿衣裳起敛容。自言本是京城女，家在虾蟆陵下住。十三学得琵琶成，名属教坊第一部。曲罢曾教善才服，妆成每被秋娘妒。五陵年少争缠头，一曲红绡不知数。钿头银篦击节碎，血色罗裙翻酒污。今年欢笑复明年，秋月春风等闲度。弟走从军阿姨死，暮去朝来颜色故。门前冷落鞍马稀，老大嫁作商人妇。商人重利轻别离，前月浮梁买茶去。去来江口守空船，绕船月明江水寒。夜深忽梦少年事，梦啼妆泪红阑干。

我闻琵琶已叹息，又闻此语重唧唧。同是天涯沦落人，相逢何必曾相识！我从去年辞帝京，谪居卧病浔阳城。浔阳地僻无音乐，终岁不闻丝竹声。住近湓江地低湿，黄芦苦竹绕宅生。其间旦暮闻何物？杜鹃啼血猿哀鸣。春江花朝秋月夜，往往取酒还独倾。岂无山歌与村笛？呕哑嘲哳难为听。今夜闻君琵琶语，如听仙乐耳暂明。莫辞更坐弹一曲，为君翻作《琵琶行》。感我此言良久立，却坐促弦弦转急。凄凄不似向前声，满座重闻皆掩泣。座中泣下谁最多？江州司马青衫湿。