原文地址
原文代码
pytorch实现1
pytorch实现2
详细讲解

EfficientNet中存在的问题

1. 训练图像尺寸大时，训练速度非常慢。train size = 512, batch = 24时，V100 out of memory
2. 在网络浅层中使用Depthwise convolutions速度会很慢。因此将原本EfficientNet中的conv1x1 and depthwise conv3x3 (MBConv)替换成conv3x3 (Fused-MBCon![请添加图片描述](https://img-blog.csdnimg.cn/fbc1131820d84da388fd075356f999f6.png) v)。但如果将所有的conv1x1 and depthwise conv3x3都替换成conv3x3后，会明显增加参数数量，降低训练速度，因此使用NAS技术去搜索两者的最佳组合。
   
3. 同等放大每个stage是次优的，因为每个stage对网络的训练速度以及参数量贡献不同。

NAS 搜索

与EfficientNet相同，但这次的NAS搜索采用了联合优化策略，联合了accuracy, parameter efficiency, training efficiency三个标准。设计空间包括

• convolutional operation types {MBConv, Fused-MBConv}
• number of layers
• kernel size {3x3, 5x5}
• expansion ration {1,4,6}
  同时随机采样1000个models，并且对每个models进行了10个epochs的训练。搜索奖励结合了模型准确率A，标准训练一个step所需要的时间S，和参数量P， $$A\cdot S^{-0.07}\cdot P^{-0.05}$$

EfficientNetV2 网络结构

与EfficientNet相比，EfficientNetV2有以下区别：

• 在浅层网络中大量运用了MBConv和新加入的fused-MBConv
• 使用了较小的expansion ratio，可以达到较小的内存访问开销
• 偏向于kernel3x3，但这需要增加层数来弥补小kernel感受野的不足
• 移除了last stride-1 stage，但是这是由于NAS搜索出来的，所以是作者的猜测可能是在参数量和访存开销的优化。

code

# 使用的是https://github.com/WZMIAOMIAO/deep-learning-for-image-processing/blob/master/pytorch_classification/Test11_efficientnetV2/model.py 中的代码！from collections import OrderedDict
from functools import partial
from typing import Callable, Optionalimport torch.nn as nn
import torch
from torch import Tensordef drop_path(x, drop_prob: float = 0., training: bool = False):"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks)."Deep Networks with Stochastic Depth", https://arxiv.org/pdf/1603.09382.pdfThis function is taken from the rwightman.It can be seen here:https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/layers/drop.py#L140"""if drop_prob == 0. or not training:return xkeep_prob = 1 - drop_probshape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNetsrandom_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)random_tensor.floor_()  # binarizeoutput = x.div(keep_prob) * random_tensorreturn outputclass DropPath(nn.Module):"""Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."Deep Networks with Stochastic Depth", https://arxiv.org/pdf/1603.09382.pdf"""def __init__(self, drop_prob=None):super(DropPath, self).__init__()self.drop_prob = drop_probdef forward(self, x):return drop_path(x, self.drop_prob, self.training)class ConvBNAct(nn.Module):def __init__(self,in_planes: int,out_planes: int,kernel_size: int = 3,stride: int = 1,groups: int = 1,norm_layer: Optional[Callable[..., nn.Module]] = None,activation_layer: Optional[Callable[..., nn.Module]] = None):super(ConvBNAct, self).__init__()padding = (kernel_size - 1) // 2if norm_layer is None:norm_layer = nn.BatchNorm2dif activation_layer is None:activation_layer = nn.SiLU  # alias Swish  (torch>=1.7)self.conv = nn.Conv2d(in_channels=in_planes,out_channels=out_planes,kernel_size=kernel_size,stride=stride,padding=padding,groups=groups,bias=False)self.bn = norm_layer(out_planes)self.act = activation_layer()def forward(self, x):result = self.conv(x)result = self.bn(result)result = self.act(result)return resultclass SqueezeExcite(nn.Module):def __init__(self,input_c: int,   # block input channelexpand_c: int,  # block expand channelse_ratio: float = 0.25):super(SqueezeExcite, self).__init__()squeeze_c = int(input_c * se_ratio)self.conv_reduce = nn.Conv2d(expand_c, squeeze_c, 1)self.act1 = nn.SiLU()  # alias Swishself.conv_expand = nn.Conv2d(squeeze_c, expand_c, 1)self.act2 = nn.Sigmoid()def forward(self, x: Tensor) -> Tensor:scale = x.mean((2, 3), keepdim=True)scale = self.conv_reduce(scale)scale = self.act1(scale)scale = self.conv_expand(scale)scale = self.act2(scale)return scale * xclass MBConv(nn.Module):def __init__(self,kernel_size: int,input_c: int,out_c: int,expand_ratio: int,stride: int,se_ratio: float,drop_rate: float,norm_layer: Callable[..., nn.Module]):super(MBConv, self).__init__()if stride not in [1, 2]:raise ValueError("illegal stride value.")self.has_shortcut = (stride == 1 and input_c == out_c)activation_layer = nn.SiLU  # alias Swishexpanded_c = input_c * expand_ratio# 在EfficientNetV2中，MBConv中不存在expansion=1的情况所以conv_pw肯定存在assert expand_ratio != 1# Point-wise expansionself.expand_conv = ConvBNAct(input_c,expanded_c,kernel_size=1,norm_layer=norm_layer,activation_layer=activation_layer)# Depth-wise convolutionself.dwconv = ConvBNAct(expanded_c,expanded_c,kernel_size=kernel_size,stride=stride,groups=expanded_c,norm_layer=norm_layer,activation_layer=activation_layer)self.se = SqueezeExcite(input_c, expanded_c, se_ratio) if se_ratio > 0 else nn.Identity()# Point-wise linear projectionself.project_conv = ConvBNAct(expanded_c,out_planes=out_c,kernel_size=1,norm_layer=norm_layer,activation_layer=nn.Identity)  # 注意这里没有激活函数，所有传入Identityself.out_channels = out_c# 只有在使用shortcut连接时才使用dropout层self.drop_rate = drop_rateif self.has_shortcut and drop_rate > 0:self.dropout = DropPath(drop_rate)def forward(self, x: Tensor) -> Tensor:result = self.expand_conv(x)result = self.dwconv(result)result = self.se(result)result = self.project_conv(result)if self.has_shortcut:if self.drop_rate > 0:result = self.dropout(result)result += xreturn resultclass FusedMBConv(nn.Module):def __init__(self,kernel_size: int,input_c: int,out_c: int,expand_ratio: int,stride: int,se_ratio: float,drop_rate: float,norm_layer: Callable[..., nn.Module]):super(FusedMBConv, self).__init__()assert stride in [1, 2]assert se_ratio == 0self.has_shortcut = stride == 1 and input_c == out_cself.drop_rate = drop_rateself.has_expansion = expand_ratio != 1activation_layer = nn.SiLU  # alias Swishexpanded_c = input_c * expand_ratio# 只有当expand ratio不等于1时才有expand convif self.has_expansion:# Expansion convolutionself.expand_conv = ConvBNAct(input_c,expanded_c,kernel_size=kernel_size,stride=stride,norm_layer=norm_layer,activation_layer=activation_layer)self.project_conv = ConvBNAct(expanded_c,out_c,kernel_size=1,norm_layer=norm_layer,activation_layer=nn.Identity)  # 注意没有激活函数else:# 当只有project_conv时的情况self.project_conv = ConvBNAct(input_c,out_c,kernel_size=kernel_size,stride=stride,norm_layer=norm_layer,activation_layer=activation_layer)  # 注意有激活函数self.out_channels = out_c# 只有在使用shortcut连接时才使用dropout层self.drop_rate = drop_rateif self.has_shortcut and drop_rate > 0:self.dropout = DropPath(drop_rate)def forward(self, x: Tensor) -> Tensor:if self.has_expansion:result = self.expand_conv(x)result = self.project_conv(result)else:result = self.project_conv(x)if self.has_shortcut:if self.drop_rate > 0:result = self.dropout(result)result += xreturn resultclass EfficientNetV2(nn.Module):def __init__(self,model_cnf: list,num_classes: int = 1000,num_features: int = 1280,dropout_rate: float = 0.2,drop_connect_rate: float = 0.2):super(EfficientNetV2, self).__init__()for cnf in model_cnf:assert len(cnf) == 8norm_layer = partial(nn.BatchNorm2d, eps=1e-3, momentum=0.1)stem_filter_num = model_cnf[0][4]self.stem = ConvBNAct(3,stem_filter_num,kernel_size=3,stride=2,norm_layer=norm_layer)  # 激活函数默认是SiLUtotal_blocks = sum([i[0] for i in model_cnf])block_id = 0blocks = []for cnf in model_cnf:repeats = cnf[0]op = FusedMBConv if cnf[-2] == 0 else MBConvfor i in range(repeats):blocks.append(op(kernel_size=cnf[1],input_c=cnf[4] if i == 0 else cnf[5],out_c=cnf[5],expand_ratio=cnf[3],stride=cnf[2] if i == 0 else 1,se_ratio=cnf[-1],drop_rate=drop_connect_rate * block_id / total_blocks,norm_layer=norm_layer))block_id += 1self.blocks = nn.Sequential(*blocks)head_input_c = model_cnf[-1][-3]head = OrderedDict()head.update({"project_conv": ConvBNAct(head_input_c,num_features,kernel_size=1,norm_layer=norm_layer)})  # 激活函数默认是SiLUhead.update({"avgpool": nn.AdaptiveAvgPool2d(1)})head.update({"flatten": nn.Flatten()})if dropout_rate > 0:head.update({"dropout": nn.Dropout(p=dropout_rate, inplace=True)})head.update({"classifier": nn.Linear(num_features, num_classes)})self.head = nn.Sequential(head)# initial weightsfor m in self.modules():if isinstance(m, nn.Conv2d):nn.init.kaiming_normal_(m.weight, mode="fan_out")if m.bias is not None:nn.init.zeros_(m.bias)elif isinstance(m, nn.BatchNorm2d):nn.init.ones_(m.weight)nn.init.zeros_(m.bias)elif isinstance(m, nn.Linear):nn.init.normal_(m.weight, 0, 0.01)nn.init.zeros_(m.bias)def forward(self, x: Tensor) -> Tensor:x = self.stem(x)x = self.blocks(x)x = self.head(x)return xdef efficientnetv2_s(num_classes: int = 1000):"""EfficientNetV2https://arxiv.org/abs/2104.00298"""# train_size: 300, eval_size: 384# repeat, kernel, stride, expansion, in_c, out_c, operator, se_ratiomodel_config = [[2, 3, 1, 1, 24, 24, 0, 0],[4, 3, 2, 4, 24, 48, 0, 0],[4, 3, 2, 4, 48, 64, 0, 0],[6, 3, 2, 4, 64, 128, 1, 0.25],[9, 3, 1, 6, 128, 160, 1, 0.25],[15, 3, 2, 6, 160, 256, 1, 0.25]]model = EfficientNetV2(model_cnf=model_config,num_classes=num_classes,dropout_rate=0.2)return modeldef efficientnetv2_m(num_classes: int = 1000):"""EfficientNetV2https://arxiv.org/abs/2104.00298"""# train_size: 384, eval_size: 480# repeat, kernel, stride, expansion, in_c, out_c, operator, se_ratiomodel_config = [[3, 3, 1, 1, 24, 24, 0, 0],[5, 3, 2, 4, 24, 48, 0, 0],[5, 3, 2, 4, 48, 80, 0, 0],[7, 3, 2, 4, 80, 160, 1, 0.25],[14, 3, 1, 6, 160, 176, 1, 0.25],[18, 3, 2, 6, 176, 304, 1, 0.25],[5, 3, 1, 6, 304, 512, 1, 0.25]]model = EfficientNetV2(model_cnf=model_config,num_classes=num_classes,dropout_rate=0.3)return modeldef efficientnetv2_l(num_classes: int = 1000):"""EfficientNetV2https://arxiv.org/abs/2104.00298"""# train_size: 384, eval_size: 480# repeat, kernel, stride, expansion, in_c, out_c, operator, se_ratiomodel_config = [[4, 3, 1, 1, 32, 32, 0, 0],[7, 3, 2, 4, 32, 64, 0, 0],[7, 3, 2, 4, 64, 96, 0, 0],[10, 3, 2, 4, 96, 192, 1, 0.25],[19, 3, 1, 6, 192, 224, 1, 0.25],[25, 3, 2, 6, 224, 384, 1, 0.25],[7, 3, 1, 6, 384, 640, 1, 0.25]]model = EfficientNetV2(model_cnf=model_config,num_classes=num_classes,dropout_rate=0.4)return modelfrom torchsummary import summary
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = efficientnetv2_l()
model = model.to(device)
summary(model, (3,256,256))

使用torchsummary输出结果:

================================================================
Total params: 118,515,272
Trainable params: 118,515,272
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.75
Forward/backward pass size (MB): 1576.33
Params size (MB): 452.10
Estimated Total Size (MB): 2029.18
----------------------------------------------------------------