这期博客我们来学习一下Efficientnet网络，属于NAS系列中最优秀的轻量级网络之一，通过NAS搜索的方式确定最佳的网络结构。之前的神经网络的宽度深度，输入图像的分辨率，是怎么获得的呢，说白了就是经验，研究人员通过无数的设计经验获得的，但是网络的发展不可能一直通过经验，如何获取最佳的结构是大家研究的问题。
本博客代码可以直接生成训练集和测试集的损失和准确率的折线图，便于写论文使用。
Efficientnet论文下载地址：https://arxiv.org/abs/1905.11946
在Efficientnet出现之前，大家都在纠结对于提高网络性能的三点要求：
1、提高网络的深度
2、提高网络的宽度
3、提高输入图片的分辨率
前两点的提高在理论上可以提升网络的学习能力，通过提升网络深度来提升网络性能的神经网络有alexnet、Vgg、resnet。通过提升网络宽度的神经网络有Googlenet系列，inception v2 inceptionv3等等。通过提升输入图像的分辨率来提升模型的性能，因为分辨率越高可以带来更细微的图像特征，使得网络可以学到更多的目标特征信息，因此可以提升网络的性能。
一举例子大家就知道这三个方面的提升到底是啥意思了，但是随着研究的深入，研究人员发现一味的提高这三项指标，并不能一直提升网络的性能。
1、网络深度的持续提升，会增加网络的学习能力，但是会出现梯度爆炸，梯度消失等问题。（可以参考Resnet诞生的原因）
2、网络宽度的持续提升，可以使得网络学到多种不同的特征，获得更为好的效果，但是对于深层次的特征无法获取，无法获取更为深层次的语义特征。
3、输入图片的分辨率越高，虽然网络可以获取更为精细化的特征，但是这样会大大增加计算量，增大对设备的要求，这显然不是我们想看到的。
因此，如何不通过人工实验的方式，获取最为理想化的网络深度、宽度、使用最合适的图片分辨率，如何平衡三者之间的关系，成为研究人员研究的问题。
在这种背景下，Efficientnet应运而生，在我的思想中，这应该是NAS搜索网络的代表作之一把，很经典的NAS搜索网络。下图是我自己画的MBConv结构，是Efficientnet中使用的，跟mobilenet v3的Inverted residual block区别不大，可以参考如下链接。
MBConv结构
MBConv具体参数
跟mobilenet v3的Inverted residual block的区别在于加入了SE（Squeeze-and-Excitation）注意力机制模块，且使用Swish激活函数。
EfficientNet作者给Efficientnet网络的8个版本，EfficientNet_b0-EfficientNet_b7
EfficientNet_b0-EfficientNet_b7的具体参数如上表所示，其中
1、input_size为输入网络的图像大小
2、width_coefficient为通道的倍率，跟通道数相乘向上取整
3、depth_coefficient深度的倍率，跟深度相乘向上取整drop_connect_rate
4、drop_connect_rate在MBConv结构中dropout层中使用
5、dropout_rate在最后一个全连接层前的dropout层中使用
论文作者给出了EfficientNet系列网络相较于当时主流的其他网络的对比，可以看出相较于ResNet-50，DenseNet等网络，EfficientNet系列网络具有明显的优势。如下图所示
Efficientnet代码
训练代码：
import torch
import math
import copy
from torch import nn, Tensor
import torchvision.models
from torch.nn import functional as F
from matplotlib import pyplot as plt
from tqdm import tqdm
from torch import nn
from torch.utils.data import DataLoader
from torchvision.transforms import transforms
from functools import partial
from collections import OrderedDict
from typing import Optional, Callable
data_transform = {
"train": transforms.Compose([transforms.RandomResizedCrop(120),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]),
"val": transforms.Compose([transforms.Resize((120, 120)), # cannot 224, must (224, 224)
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])}
train_data = torchvision.datasets.ImageFolder(root = "./data/train" , transform = data_transform["train"])
traindata = DataLoader(dataset=train_data, batch_size = 32 , shuffle=True, num_workers=0) # 将训练数据以每次32张图片的形式抽出进行训练
test_data = torchvision.datasets.ImageFolder(root = "./data/val" , transform = data_transform["val"])
train_size = len(train_data) # 训练集的长度
test_size = len(test_data) # 测试集的长度
print(train_size) #输出训练集长度看一下，相当于看看有几张图片
print(test_size) #输出测试集长度看一下，相当于看看有几张图片
testdata = DataLoader(dataset=test_data, batch_size = 32, shuffle=True, num_workers=0) # 将训练数据以每次32张图片的形式抽出进行测试
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print("using {} device.".format(device))
def _make_divisible(ch, divisor=8, min_ch=None):
"""
This function is taken from the original tf repo.
It ensures that all layers have a channel number that is divisible by 8
It can be seen here:
https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
"""
if min_ch is None:
min_ch = divisor
new_ch = max(min_ch, int(ch + divisor / 2) // divisor * divisor)
# Make sure that round down does not go down by more than 10%.
if new_ch < 0.9 * ch:
new_ch += divisor
return new_ch
def 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.pdf
This 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 x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_() # binarize
output = x.div(keep_prob) * random_tensor
return output
class 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_prob
def forward(self, x):
return drop_path(x, self.drop_prob, self.training)
class ConvBNActivation(nn.Sequential):
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):
padding = (kernel_size - 1) // 2
if norm_layer is None:
norm_layer = nn.BatchNorm2d
if activation_layer is None:
activation_layer = nn.SiLU # alias Swish (torch>=1.7)
super(ConvBNActivation, self).__init__(nn.Conv2d(in_channels=in_planes,
out_channels=out_planes,
kernel_size=kernel_size,
stride=stride,
padding=padding,
groups=groups,
bias=False),
norm_layer(out_planes),
activation_layer())
class SqueezeExcitation(nn.Module):
def __init__(self,
input_c: int, # block input channel
expand_c: int, # block expand channel
squeeze_factor: int = 4):
super(SqueezeExcitation, self).__init__()
squeeze_c = input_c // squeeze_factor
self.fc1 = nn.Conv2d(expand_c, squeeze_c, 1)
self.ac1 = nn.SiLU() # alias Swish
self.fc2 = nn.Conv2d(squeeze_c, expand_c, 1)
self.ac2 = nn.Sigmoid()
def forward(self, x: Tensor) -> Tensor:
scale = F.adaptive_avg_pool2d(x, output_size=(1, 1))
scale = self.fc1(scale)
scale = self.ac1(scale)
scale = self.fc2(scale)
scale = self.ac2(scale)
return scale * x
class InvertedResidualConfig:
# kernel_size, in_channel, out_channel, exp_ratio, strides, use_SE, drop_connect_rate
def __init__(self,
kernel: int, # 3 or 5
input_c: int,
out_c: int,
expanded_ratio: int, # 1 or 6
stride: int, # 1 or 2
use_se: bool, # True
drop_rate: float,
index: str, # 1a, 2a, 2b, ...
width_coefficient: float):
self.input_c = self.adjust_channels(input_c, width_coefficient)
self.kernel = kernel
self.expanded_c = self.input_c * expanded_ratio
self.out_c = self.adjust_channels(out_c, width_coefficient)
self.use_se = use_se
self.stride = stride
self.drop_rate = drop_rate
self.index = index
@staticmethod
def adjust_channels(channels: int, width_coefficient: float):
return _make_divisible(channels * width_coefficient, 8)
class InvertedResidual(nn.Module):
def __init__(self,
cnf: InvertedResidualConfig,
norm_layer: Callable[..., nn.Module]):
super(InvertedResidual, self).__init__()
if cnf.stride not in [1, 2]:
raise ValueError("illegal stride value.")
self.use_res_connect = (cnf.stride == 1 and cnf.input_c == cnf.out_c)
layers = OrderedDict()
activation_layer = nn.SiLU # alias Swish
# expand
if cnf.expanded_c != cnf.input_c:
layers.update({"expand_conv": ConvBNActivation(cnf.input_c,
cnf.expanded_c,
kernel_size=1,
norm_layer=norm_layer,
activation_layer=activation_layer)})
# depthwise
layers.update({"dwconv": ConvBNActivation(cnf.expanded_c,
cnf.expanded_c,
kernel_size=cnf.kernel,
stride=cnf.stride,
groups=cnf.expanded_c,
norm_layer=norm_layer,
activation_layer=activation_layer)})
if cnf.use_se:
layers.update({"se": SqueezeExcitation(cnf.input_c,
cnf.expanded_c)})
# project
layers.update({"project_conv": ConvBNActivation(cnf.expanded_c,
cnf.out_c,
kernel_size=1,
norm_layer=norm_layer,
activation_layer=nn.Identity)})
self.block = nn.Sequential(layers)
self.out_channels = cnf.out_c
self.is_strided = cnf.stride > 1
# 只有在使用shortcut连接时才使用dropout层
if self.use_res_connect and cnf.drop_rate > 0:
self.dropout = DropPath(cnf.drop_rate)
else:
self.dropout = nn.Identity()
def forward(self, x: Tensor) -> Tensor:
result = self.block(x)
result = self.dropout(result)
if self.use_res_connect:
result += x
return result
class EfficientNet(nn.Module):
def __init__(self,
width_coefficient: float,
depth_coefficient: float,
num_classes: int = 1000,
dropout_rate: float = 0.2,
drop_connect_rate: float = 0.2,
block: Optional[Callable[..., nn.Module]] = None,
norm_layer: Optional[Callable[..., nn.Module]] = None
):
super(EfficientNet, self).__init__()
# kernel_size, in_channel, out_channel, exp_ratio, strides, use_SE, drop_connect_rate, repeats
default_cnf = [[3, 32, 16, 1, 1, True, drop_connect_rate, 1],
[3, 16, 24, 6, 2, True, drop_connect_rate, 2],
[5, 24, 40, 6, 2, True, drop_connect_rate, 2],
[3, 40, 80, 6, 2, True, drop_connect_rate, 3],
[5, 80, 112, 6, 1, True, drop_connect_rate, 3],
[5, 112, 192, 6, 2, True, drop_connect_rate, 4],
[3, 192, 320, 6, 1, True, drop_connect_rate, 1]]
def round_repeats(repeats):
"""Round number of repeats based on depth multiplier."""
return int(math.ceil(depth_coefficient * repeats))
if block is None:
block = InvertedResidual
if norm_layer is None:
norm_layer = partial(nn.BatchNorm2d, eps=1e-3, momentum=0.1)
adjust_channels = partial(InvertedResidualConfig.adjust_channels,
width_coefficient=width_coefficient)
# build inverted_residual_setting
bneck_conf = partial(InvertedResidualConfig,
width_coefficient=width_coefficient)
b = 0
num_blocks = float(sum(round_repeats(i[-1]) for i in default_cnf))
inverted_residual_setting = []
for stage, args in enumerate(default_cnf):
cnf = copy.copy(args)
for i in range(round_repeats(cnf.pop(-1))):
if i > 0:
# strides equal 1 except first cnf
cnf[-3] = 1 # strides
cnf[1] = cnf[2] # input_channel equal output_channel
cnf[-1] = args[-2] * b / num_blocks # update dropout ratio
index = str(stage + 1) + chr(i + 97) # 1a, 2a, 2b, ...
inverted_residual_setting.append(bneck_conf(*cnf, index))
b += 1
# create layers
layers = OrderedDict()
# first conv
layers.update({"stem_conv": ConvBNActivation(in_planes=3,
out_planes=adjust_channels(32),
kernel_size=3,
stride=2,
norm_layer=norm_layer)})
# building inverted residual blocks
for cnf in inverted_residual_setting:
layers.update({cnf.index: block(cnf, norm_layer)})
# build top
last_conv_input_c = inverted_residual_setting[-1].out_c
last_conv_output_c = adjust_channels(1280)
layers.update({"top": ConvBNActivation(in_planes=last_conv_input_c,
out_planes=last_conv_output_c,
kernel_size=1,
norm_layer=norm_layer)})
self.features = nn.Sequential(layers)
self.avgpool = nn.AdaptiveAvgPool2d(1)
classifier = []
if dropout_rate > 0:
classifier.append(nn.Dropout(p=dropout_rate, inplace=True))
classifier.append(nn.Linear(last_conv_output_c, num_classes))
self.classifier = nn.Sequential(*classifier)
# initial weights
for 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_impl(self, x: Tensor) -> Tensor:
x = self.features(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.classifier(x)
return x
def forward(self, x: Tensor) -> Tensor:
return self._forward_impl(x)
def efficientnet_b0(num_classes=1000):
# input image size 224x224
return EfficientNet(width_coefficient=1.0,
depth_coefficient=1.0,
dropout_rate=0.2,
num_classes=num_classes)
def efficientnet_b1(num_classes=1000):
# input image size 240x240
return EfficientNet(width_coefficient=1.0,
depth_coefficient=1.1,
dropout_rate=0.2,
num_classes=num_classes)
def efficientnet_b2(num_classes=1000):
# input image size 260x260
return EfficientNet(width_coefficient=1.1,
depth_coefficient=1.2,
dropout_rate=0.3,
num_classes=num_classes)
def efficientnet_b3(num_classes=1000):
# input image size 300x300
return EfficientNet(width_coefficient=1.2,
depth_coefficient=1.4,
dropout_rate=0.3,
num_classes=num_classes)
def efficientnet_b4(num_classes=1000):
# input image size 380x380
return EfficientNet(width_coefficient=1.4,
depth_coefficient=1.8,
dropout_rate=0.4,
num_classes=num_classes)
def efficientnet_b5(num_classes=1000):
# input image size 456x456
return EfficientNet(width_coefficient=1.6,
depth_coefficient=2.2,
dropout_rate=0.4,
num_classes=num_classes)
def efficientnet_b6(num_classes=1000):
# input image size 528x528
return EfficientNet(width_coefficient=1.8,
depth_coefficient=2.6,
dropout_rate=0.5,
num_classes=num_classes)
def efficientnet_b7(num_classes=1000):
# input image size 600x600
return EfficientNet(width_coefficient=2.0,
depth_coefficient=3.1,
dropout_rate=0.5,
num_classes=num_classes)
efficientnet = efficientnet_b0(num_classes = 2) #将模型命名为efficientnet，这里num_classes是数据集的种类，我用的猫狗数据集两类，所以等于2 你设置成你数据集的种类即可
#上面用的是efficientnet_b0，如果想用efficientnet其他系列的，直接把上面的efficientnet_b其他系列就行即可
efficientnet.to(device)
print(efficientnet.to(device)) #输出模型结构
test1 = torch.ones(64, 3, 120, 120) # 测试一下输出的形状大小 输入一个64,3,120,120的向量
test1 = efficientnet(test1.to(device)) #将向量打入神经网络进行测试
print(test1.shape) #查看输出的结果
epoch = 1 # 迭代次数即训练次数
learning = 0.0001 # 学习率
optimizer = torch.optim.Adam(efficientnet.parameters(), lr=learning) # 使用Adam优化器-写论文的话可以具体查一下这个优化器的原理
loss = nn.CrossEntropyLoss() # 损失计算方式，交叉熵损失函数
train_loss_all = [] # 存放训练集损失的数组
train_accur_all = [] # 存放训练集准确率的数组
test_loss_all = [] # 存放测试集损失的数组
test_accur_all = [] # 存放测试集准确率的数组
for i in range(epoch): #开始迭代
train_loss = 0 #训练集的损失初始设为0
train_num = 0.0 #
train_accuracy = 0.0 #训练集的准确率初始设为0
efficientnet.train() #将模型设置成 训练模式
train_bar = tqdm(traindata) #用于进度条显示，没啥实际用处
for step, data in enumerate(train_bar): #开始迭代跑， enumerate这个函数不懂可以查查，将训练集分为 data是序号，data是数据
img, target = data #将data 分位 img图片，target标签
optimizer.zero_grad() # 清空历史梯度
outputs = efficientnet(img.to(device)) # 将图片打入网络进行训练,outputs是输出的结果
loss1 = loss(outputs, target.to(device)) # 计算神经网络输出的结果outputs与图片真实标签target的差别-这就是我们通常情况下称为的损失
outputs = torch.argmax(outputs, 1) #会输出10个值，最大的值就是我们预测的结果 求最大值
loss1.backward() #神经网络反向传播
optimizer.step() #梯度优化 用上面的abam优化
train_loss = train_loss + loss1.item() #将所有损失的绝对值加起来
accuracy = torch.sum(outputs == target.to(device)) #outputs == target的 即使预测正确的，统计预测正确的个数,从而计算准确率
train_accuracy = train_accuracy + accuracy #求训练集的准确率
train_num += img.size(0) #
print("epoch：{} ， train-Loss：{} , train-accuracy：{}".format(i + 1, train_loss / train_num, #输出训练情况
train_accuracy / train_num))
train_loss_all.append(train_loss / train_num) #将训练的损失放到一个列表里 方便后续画图
train_accur_all.append(train_accuracy.double().item() / train_num)#训练集的准确率
test_loss = 0 #同上 测试损失
test_accuracy = 0.0 #测试准确率
test_num = 0
efficientnet.eval() #将模型调整为测试模型
with torch.no_grad(): #清空历史梯度，进行测试 与训练最大的区别是测试过程中取消了反向传播
test_bar = tqdm(testdata)
for data in test_bar:
img, target = data
outputs = efficientnet(img.to(device))
loss2 = loss(outputs, target.to(device)).cpu()
outputs = torch.argmax(outputs, 1)
test_loss = test_loss + loss2.item()
accuracy = torch.sum(outputs == target.to(device))
test_accuracy = test_accuracy + accuracy
test_num += img.size(0)
print("test-Loss：{} , test-accuracy：{}".format(test_loss / test_num, test_accuracy / test_num))
test_loss_all.append(test_loss / test_num)
test_accur_all.append(test_accuracy.double().item() / test_num)
#下面的是画图过程，将上述存放的列表 画出来即可，分别画出训练集和测试集的损失和准确率图
plt.figure(figsize=(12, 4))
plt.subplot(1, 2, 1)
plt.plot(range(epoch), train_loss_all,
"ro-", label="Train loss")
plt.plot(range(epoch), test_loss_all,
"bs-", label="test loss")
plt.legend()
plt.xlabel("epoch")
plt.ylabel("Loss")
plt.subplot(1, 2, 2)
plt.plot(range(epoch), train_accur_all,
"ro-", label="Train accur")
plt.plot(range(epoch), test_accur_all,
"bs-", label="test accur")
plt.xlabel("epoch")
plt.ylabel("acc")
plt.legend()
plt.show()
torch.save(efficientnet, "efficientnet.pth")
print("模型已保存")
预测代码：
import torch
from PIL import Image
from torch import nn
from torchvision.transforms import transforms
from typing import Callable, List, Optional
from torch import nn, Tensor
from torch.nn import functional as F
image_path = "1.jpg"#相对路径 导入图片
trans = transforms.Compose([transforms.Resize((120 , 120)),
transforms.ToTensor()]) #将图片缩放为跟训练集图片的大小一样 方便预测，且将图片转换为张量
image = Image.open(image_path) #打开图片
# print(image) #输出图片 看看图片格式
image = image.convert("RGB") #将图片转换为RGB格式
image = trans(image) #上述的缩放和转张量操作在这里实现
# print(image) #查看转换后的样子
image = torch.unsqueeze(image, dim=0) #将图片维度扩展一维
classes = ["cat" , "dog" ] #预测种类，我这里用的猫狗数据集，所以是这两种，你调成你的种类即可
def _make_divisible(ch, divisor=8, min_ch=None):
"""
This function is taken from the original tf repo.
It ensures that all layers have a channel number that is divisible by 8
It can be seen here:
https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
"""
if min_ch is None:
min_ch = divisor
new_ch = max(min_ch, int(ch + divisor / 2) // divisor * divisor)
# Make sure that round down does not go down by more than 10%.
if new_ch < 0.9 * ch:
new_ch += divisor
return new_ch
def 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.pdf
This 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 x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_() # binarize
output = x.div(keep_prob) * random_tensor
return output
class 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_prob
def forward(self, x):
return drop_path(x, self.drop_prob, self.training)
class ConvBNActivation(nn.Sequential):
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):
padding = (kernel_size - 1) // 2
if norm_layer is None:
norm_layer = nn.BatchNorm2d
if activation_layer is None:
activation_layer = nn.SiLU # alias Swish (torch>=1.7)
super(ConvBNActivation, self).__init__(nn.Conv2d(in_channels=in_planes,
out_channels=out_planes,
kernel_size=kernel_size,
stride=stride,
padding=padding,
groups=groups,
bias=False),
norm_layer(out_planes),
activation_layer())
class SqueezeExcitation(nn.Module):
def __init__(self,
input_c: int, # block input channel
expand_c: int, # block expand channel
squeeze_factor: int = 4):
super(SqueezeExcitation, self).__init__()
squeeze_c = input_c // squeeze_factor
self.fc1 = nn.Conv2d(expand_c, squeeze_c, 1)
self.ac1 = nn.SiLU() # alias Swish
self.fc2 = nn.Conv2d(squeeze_c, expand_c, 1)
self.ac2 = nn.Sigmoid()
def forward(self, x: Tensor) -> Tensor:
scale = F.adaptive_avg_pool2d(x, output_size=(1, 1))
scale = self.fc1(scale)
scale = self.ac1(scale)
scale = self.fc2(scale)
scale = self.ac2(scale)
return scale * x
class InvertedResidualConfig:
# kernel_size, in_channel, out_channel, exp_ratio, strides, use_SE, drop_connect_rate
def __init__(self,
kernel: int, # 3 or 5
input_c: int,
out_c: int,
expanded_ratio: int, # 1 or 6
stride: int, # 1 or 2
use_se: bool, # True
drop_rate: float,
index: str, # 1a, 2a, 2b, ...
width_coefficient: float):
self.input_c = self.adjust_channels(input_c, width_coefficient)
self.kernel = kernel
self.expanded_c = self.input_c * expanded_ratio
self.out_c = self.adjust_channels(out_c, width_coefficient)
self.use_se = use_se
self.stride = stride
self.drop_rate = drop_rate
self.index = index
@staticmethod
def adjust_channels(channels: int, width_coefficient: float):
return _make_divisible(channels * width_coefficient, 8)
class InvertedResidual(nn.Module):
def __init__(self,
cnf: InvertedResidualConfig,
norm_layer: Callable[..., nn.Module]):
super(InvertedResidual, self).__init__()
if cnf.stride not in [1, 2]:
raise ValueError("illegal stride value.")
self.use_res_connect = (cnf.stride == 1 and cnf.input_c == cnf.out_c)
layers = OrderedDict()
activation_layer = nn.SiLU # alias Swish
# expand
if cnf.expanded_c != cnf.input_c:
layers.update({"expand_conv": ConvBNActivation(cnf.input_c,
cnf.expanded_c,
kernel_size=1,
norm_layer=norm_layer,
activation_layer=activation_layer)})
# depthwise
layers.update({"dwconv": ConvBNActivation(cnf.expanded_c,
cnf.expanded_c,
kernel_size=cnf.kernel,
stride=cnf.stride,
groups=cnf.expanded_c,
norm_layer=norm_layer,
activation_layer=activation_layer)})
if cnf.use_se:
layers.update({"se": SqueezeExcitation(cnf.input_c,
cnf.expanded_c)})
# project
layers.update({"project_conv": ConvBNActivation(cnf.expanded_c,
cnf.out_c,
kernel_size=1,
norm_layer=norm_layer,
activation_layer=nn.Identity)})
self.block = nn.Sequential(layers)
self.out_channels = cnf.out_c
self.is_strided = cnf.stride > 1
# 只有在使用shortcut连接时才使用dropout层
if self.use_res_connect and cnf.drop_rate > 0:
self.dropout = DropPath(cnf.drop_rate)
else:
self.dropout = nn.Identity()
def forward(self, x: Tensor) -> Tensor:
result = self.block(x)
result = self.dropout(result)
if self.use_res_connect:
result += x
return result
class EfficientNet(nn.Module):
def __init__(self,
width_coefficient: float,
depth_coefficient: float,
num_classes: int = 1000,
dropout_rate: float = 0.2,
drop_connect_rate: float = 0.2,
block: Optional[Callable[..., nn.Module]] = None,
norm_layer: Optional[Callable[..., nn.Module]] = None
):
super(EfficientNet, self).__init__()
# kernel_size, in_channel, out_channel, exp_ratio, strides, use_SE, drop_connect_rate, repeats
default_cnf = [[3, 32, 16, 1, 1, True, drop_connect_rate, 1],
[3, 16, 24, 6, 2, True, drop_connect_rate, 2],
[5, 24, 40, 6, 2, True, drop_connect_rate, 2],
[3, 40, 80, 6, 2, True, drop_connect_rate, 3],
[5, 80, 112, 6, 1, True, drop_connect_rate, 3],
[5, 112, 192, 6, 2, True, drop_connect_rate, 4],
[3, 192, 320, 6, 1, True, drop_connect_rate, 1]]
def round_repeats(repeats):
"""Round number of repeats based on depth multiplier."""
return int(math.ceil(depth_coefficient * repeats))
if block is None:
block = InvertedResidual
if norm_layer is None:
norm_layer = partial(nn.BatchNorm2d, eps=1e-3, momentum=0.1)
adjust_channels = partial(InvertedResidualConfig.adjust_channels,
width_coefficient=width_coefficient)
# build inverted_residual_setting
bneck_conf = partial(InvertedResidualConfig,
width_coefficient=width_coefficient)
b = 0
num_blocks = float(sum(round_repeats(i[-1]) for i in default_cnf))
inverted_residual_setting = []
for stage, args in enumerate(default_cnf):
cnf = copy.copy(args)
for i in range(round_repeats(cnf.pop(-1))):
if i > 0:
# strides equal 1 except first cnf
cnf[-3] = 1 # strides
cnf[1] = cnf[2] # input_channel equal output_channel
cnf[-1] = args[-2] * b / num_blocks # update dropout ratio
index = str(stage + 1) + chr(i + 97) # 1a, 2a, 2b, ...
inverted_residual_setting.append(bneck_conf(*cnf, index))
b += 1
# create layers
layers = OrderedDict()
# first conv
layers.update({"stem_conv": ConvBNActivation(in_planes=3,
out_planes=adjust_channels(32),
kernel_size=3,
stride=2,
norm_layer=norm_layer)})
# building inverted residual blocks
for cnf in inverted_residual_setting:
layers.update({cnf.index: block(cnf, norm_layer)})
# build top
last_conv_input_c = inverted_residual_setting[-1].out_c
last_conv_output_c = adjust_channels(1280)
layers.update({"top": ConvBNActivation(in_planes=last_conv_input_c,
out_planes=last_conv_output_c,
kernel_size=1,
norm_layer=norm_layer)})
self.features = nn.Sequential(layers)
self.avgpool = nn.AdaptiveAvgPool2d(1)
classifier = []
if dropout_rate > 0:
classifier.append(nn.Dropout(p=dropout_rate, inplace=True))
classifier.append(nn.Linear(last_conv_output_c, num_classes))
self.classifier = nn.Sequential(*classifier)
# initial weights
for 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_impl(self, x: Tensor) -> Tensor:
x = self.features(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.classifier(x)
return x
def forward(self, x: Tensor) -> Tensor:
return self._forward_impl(x)
def efficientnet_b0(num_classes=1000):
# input image size 224x224
return EfficientNet(width_coefficient=1.0,
depth_coefficient=1.0,
dropout_rate=0.2,
num_classes=num_classes)
def efficientnet_b1(num_classes=1000):
# input image size 240x240
return EfficientNet(width_coefficient=1.0,
depth_coefficient=1.1,
dropout_rate=0.2,
num_classes=num_classes)
def efficientnet_b2(num_classes=1000):
# input image size 260x260
return EfficientNet(width_coefficient=1.1,
depth_coefficient=1.2,
dropout_rate=0.3,
num_classes=num_classes)
def efficientnet_b3(num_classes=1000):
# input image size 300x300
return EfficientNet(width_coefficient=1.2,
depth_coefficient=1.4,
dropout_rate=0.3,
num_classes=num_classes)
def efficientnet_b4(num_classes=1000):
# input image size 380x380
return EfficientNet(width_coefficient=1.4,
depth_coefficient=1.8,
dropout_rate=0.4,
num_classes=num_classes)
def efficientnet_b5(num_classes=1000):
# input image size 456x456
return EfficientNet(width_coefficient=1.6,
depth_coefficient=2.2,
dropout_rate=0.4,
num_classes=num_classes)
def efficientnet_b6(num_classes=1000):
# input image size 528x528
return EfficientNet(width_coefficient=1.8,
depth_coefficient=2.6,
dropout_rate=0.5,
num_classes=num_classes)
def efficientnet_b7(num_classes=1000):
# input image size 600x600
return EfficientNet(width_coefficient=2.0,
depth_coefficient=3.1,
dropout_rate=0.5,
num_classes=num_classes)
#以上是神经网络结构，因为读取了模型之后代码还得知道神经网络的结构才能进行预测
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") #将代码放入GPU进行训练
print("using {} device.".format(device))
model = torch.load("efficientnet.pth") #读取模型
model.eval() #关闭梯度，将模型调整为测试模式
with torch.no_grad(): #梯度清零
outputs = model(image.to(device)) #将图片打入神经网络进行测试
# print(model) #输出模型结构
# print(outputs) #输出预测的张量数组
ans = (outputs.argmax(1)).item() #最大的值即为预测结果，找出最大值在数组中的序号，
# 对应找其在种类中的序号即可然后输出即为其种类
print("该图片的种类为：",classes[ans])
# print(classes[ans])网络结构搭建部分注释的很详细，有问题朋友欢迎在评论区指出，感谢！
不懂我代码使用方法的可以看看我之前开源的代码，更为详细：手撕Resnet卷积神经网络-pytorch-详细注释版（可以直接替换自己数据集）-直接放置自己的数据集就能直接跑。跑的代码有问题的可以在评论区指出，看到了会回复。训练代码和预测代码均有。_pytorch更换数据集需要改代码吗_小馨馨的小翟的博客-CSDN博客
本代码使用的数据集是猫狗数据集，已经分好训练集和测试集了，下面给出数据集的下载链接。
链接：https://pan.baidu.com/s/1_gUznMQnzI0UhvsV7wPgzw
提取码：3ixd
Efficientnet 代码下载链接如下：
链接：https://pan.baidu.com/s/1czqLCOBtcxkwBgIiwq83tg
提取码：x15v
MBConv结构图ppt可修改下载链接：https://pan.baidu.com/s/1T1FNRycPgIaBTjsHE25kzA
提取码：05r2
3646
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
