第P9周：YOLOv5-backbone模块实现
文为「365天深度学习训练营」内部文章
参考本文所写记录性文章，请在文章开头带上「👉声明」
本次我将利用YOLOv5算法中的Backbone模块搭建网络，后续理论部分介绍将在语雀以及公众号（K同学啊）中详细展开，本次内容除了网络结构部分外，其余部分均与上周相同。
YOLOv5是目标检测算法，是否可以尝试将其网络结构用在目标识别上，或进行改进形成一个全新的算法（类似之前介绍过的VGG1-6）。如果效果不错的话，还可以搞一篇期刊文章出来~
分享一张我自己绘制的YOLOv5_6.0版本的算法框架图，希望它可以有助于你完成本次探索~
🏡 我的环境：
- 语言环境：Python3.12
- 编译器：VS Code
- 数据集：天气识别数据集
- 深度学习环境：Pytorch
- torch==2.5.1+mps
- torchvision==0.20.1
一、 前期准备
1. 设置GPU
如果设备上支持GPU就使用GPU,否则使用CPU
import torch
import torch.nn as nn
from torchvision import transforms, datasets
import PIL,pathlib,warnings
warnings.filterwarnings("ignore") #忽略警告信息
device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
device
device(type='mps')
2. 导入数据
data_dir = './weather_photos/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("/")[1] for path in data_paths]
classeNames
['cloudy', 'rain', 'shine', 'sunrise']
# 关于transforms.Compose的更多介绍可以参考：https://blog.youkuaiyun.com/qq_38251616/article/details/124878863
train_transforms = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
# transforms.RandomHorizontalFlip(), # 随机水平翻转
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
test_transform = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
total_data = datasets.ImageFolder("./weather_photos/",transform=train_transforms)
total_data
Dataset ImageFolder
Number of datapoints: 1125
Root location: ./weather_photos/
StandardTransform
Transform: Compose(
Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=True)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
total_data.class_to_idx
{'cloudy': 0, 'rain': 1, 'shine': 2, 'sunrise': 3}
3. 划分数据集
train_size = int(0.8 * len(total_data))
test_size = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
train_dataset, test_dataset
(<torch.utils.data.dataset.Subset at 0x317bfb740>,
<torch.utils.data.dataset.Subset at 0x317bfa9c0>)
batch_size = 4
train_dl = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
for X, y in test_dl:
print("Shape of X [N, C, H, W]: ", X.shape)
print("Shape of y: ", y.shape, y.dtype)
break
Shape of X [N, C, H, W]: torch.Size([4, 3, 224, 224])
Shape of y: torch.Size([4]) torch.int64
二、搭建模型
1. 搭建模型
import torch.nn.functional as F
def autopad(k, p=None): # kernel, padding
# Pad to 'same'
if p is None:
# # k 是 int 整数则除以2, 若干的整数值则循环整除
p = k // 2 if isinstance(k, int) else [x // 2 for x in k] # auto-pad
return p
class Conv(nn.Module):
# Standard convolution
def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True): # ch_in, ch_out, kernel, stride, padding, groups
"""
:param c1: 输入的channel值
:param c2: 输出的channel值
:param k: 卷积的kernel_size
:param s: 卷积的stride
:param p: 卷积的padding 一般是None
:param act: 激活函数类型 True就是SiLU(), False就是不使用激活函数
:param g: 卷积的groups数 =1就是普通的卷积 >1就是深度可分离卷积
"""
super().__init__()
self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)
self.bn = nn.BatchNorm2d(c2)
# 若act=True, 则激活, act=False, 不激活
self.act = nn.SiLU() if act is True else (act if isinstance(act, nn.Module) else nn.Identity())
def forward(self, x):
return self.act(self.bn(self.conv(x)))
class Bottleneck(nn.Module):
# Standard bottleneck
"""
:param c1: 整个Bottleneck的输入channel
:param c2: 整个Bottleneck的输出channel
:param e: expansion ratio c2*e 就是第一个卷积的输出channel=第二个卷积的输入channel
:param shortcut: bool Bottleneck中是否有shortcut，默认True
:param g: Bottleneck中的3x3卷积类型 =1普通卷积 >1深度可分离卷积
"""
def __init__(self, c1, c2, shortcut=True, g=1, e=0.5): # ch_in, ch_out, shortcut, groups, expansion
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = Conv(c1, c_, 1, 1)
self.cv2 = Conv(c_, c2, 3, 1, g=g)
self.add = shortcut and c1 == c2
def forward(self, x):
return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
class C3(nn.Module):
# CSP Bottleneck with 3 convolutions
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5): # ch_in, ch_out, number, shortcut, groups, expansion
"""
:param c1: 整个 C3 的输入channel
:param c2: 整个 C3 的输出channel
:param n: 有n个Bottleneck
:param shortcut: bool Bottleneck中是否有shortcut，默认True
:param g: C3中的3x3卷积类型 =1普通卷积 >1深度可分离卷积
:param e: expansion ratio
"""
super().__init__()
c_ = int(c2 * e) # hidden channels
self.cv1 = Conv(c1, c_, 1, 1)
self.cv2 = Conv(c1, c_, 1, 1)
self.cv3 = Conv(2 * c_, c2, 1) # act=FReLU(c2)
self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
def forward(self, x):
return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), dim=1))
class SPPF(nn.Module):
# Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher
def __init__(self, c1, c2, k=5): # equivalent to SPP(k=(5, 9, 13))
super().__init__()
c_ = c1 // 2 # hidden channels
self.cv1 = Conv(c1, c_, 1, 1)
self.cv2 = Conv(c_ * 4, c2, 1, 1)
self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
def forward(self, x):
x = self.cv1(x)
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress torch 1.9.0 max_pool2d() warning
y1 = self.m(x)
y2 = self.m(y1)
return self.cv2(torch.cat([x, y1, y2, self.m(y2)], 1))
"""
这个是YOLOv5, 6.0版本的主干网络，这里进行复现
（注：有部分删改，详细讲解将在后续进行展开）
"""
class YOLOv5_backbone(nn.Module):
def __init__(self):
super(YOLOv5_backbone, self).__init__()
self.Conv_1 = Conv(3, 64, 3, 2, 2)
self.Conv_2 = Conv(64, 128, 3, 2)
self.C3_3 = C3(128,128)
self.Conv_4 = Conv(128, 256, 3, 2)
self.C3_5 = C3(256,256)
self.Conv_6 = Conv(256, 512, 3, 2)
self.C3_7 = C3(512,512)
self.Conv_8 = Conv(512, 1024, 3, 2)
self.C3_9 = C3(1024, 1024)
self.SPPF = SPPF(1024, 1024, 5)
# 全连接网络层，用于分类
self.classifier = nn.Sequential(
nn.Linear(in_features=65536, out_features=100),
nn.ReLU(),
nn.Linear(in_features=100, out_features=4)
)
def forward(self, x):
x = self.Conv_1(x)
x = self.Conv_2(x)
x = self.C3_3(x)
x = self.Conv_4(x)
x = self.C3_5(x)
x = self.Conv_6(x)
x = self.C3_7(x)
x = self.Conv_8(x)
x = self.C3_9(x)
x = self.SPPF(x)
x = torch.flatten(x, start_dim=1)
x = self.classifier(x)
return x
print("Using {} device".format(device))
model = YOLOv5_backbone().to(device)
model
Using mps device
YOLOv5_backbone(
(Conv_1): Conv(
(conv): Conv2d(3, 64, kernel_size=(3, 3), stride=(2, 2), padding=(2, 2), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(Conv_2): Conv(
(conv): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(C3_3): C3(
(cv1): Conv(
(conv): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(128, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv3): Conv(
(conv): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(m): Sequential(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
)
)
)
(Conv_4): Conv(
(conv): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(C3_5): C3(
(cv1): Conv(
(conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv3): Conv(
(conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(m): Sequential(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
)
)
)
(Conv_6): Conv(
(conv): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(C3_7): C3(
(cv1): Conv(
(conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv3): Conv(
(conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(m): Sequential(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
)
)
)
(Conv_8): Conv(
(conv): Conv2d(512, 1024, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(C3_9): C3(
(cv1): Conv(
(conv): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv3): Conv(
(conv): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(m): Sequential(
(0): Bottleneck(
(cv1): Conv(
(conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
)
)
)
(SPPF): SPPF(
(cv1): Conv(
(conv): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(cv2): Conv(
(conv): Conv2d(2048, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(act): SiLU()
)
(m): MaxPool2d(kernel_size=5, stride=1, padding=2, dilation=1, ceil_mode=False)
)
(classifier): Sequential(
(0): Linear(in_features=65536, out_features=100, bias=True)
(1): ReLU()
(2): Linear(in_features=100, out_features=4, bias=True)
)
)
2. 查看模型详情
# 统计模型参数量以及其他指标
from torchinfo import summary
model = model.to("cpu")
batch_size = 32
summary(model, input_size = (batch_size, 3, 224, 224))
===============================================================================================
Layer (type:depth-idx) Output Shape Param #
===============================================================================================
YOLOv5_backbone [32, 4] --
├─Conv: 1-1 [32, 64, 113, 113] --
│ └─Conv2d: 2-1 [32, 64, 113, 113] 1,728
│ └─BatchNorm2d: 2-2 [32, 64, 113, 113] 128
│ └─SiLU: 2-3 [32, 64, 113, 113] --
├─Conv: 1-2 [32, 128, 57, 57] --
│ └─Conv2d: 2-4 [32, 128, 57, 57] 73,728
│ └─BatchNorm2d: 2-5 [32, 128, 57, 57] 256
│ └─SiLU: 2-6 [32, 128, 57, 57] --
├─C3: 1-3 [32, 128, 57, 57] --
│ └─Conv: 2-7 [32, 64, 57, 57] --
│ │ └─Conv2d: 3-1 [32, 64, 57, 57] 8,192
│ │ └─BatchNorm2d: 3-2 [32, 64, 57, 57] 128
│ │ └─SiLU: 3-3 [32, 64, 57, 57] --
│ └─Sequential: 2-8 [32, 64, 57, 57] --
│ │ └─Bottleneck: 3-4 [32, 64, 57, 57] 41,216
│ └─Conv: 2-9 [32, 64, 57, 57] --
│ │ └─Conv2d: 3-5 [32, 64, 57, 57] 8,192
│ │ └─BatchNorm2d: 3-6 [32, 64, 57, 57] 128
│ │ └─SiLU: 3-7 [32, 64, 57, 57] --
│ └─Conv: 2-10 [32, 128, 57, 57] --
│ │ └─Conv2d: 3-8 [32, 128, 57, 57] 16,384
│ │ └─BatchNorm2d: 3-9 [32, 128, 57, 57] 256
│ │ └─SiLU: 3-10 [32, 128, 57, 57] --
├─Conv: 1-4 [32, 256, 29, 29] --
│ └─Conv2d: 2-11 [32, 256, 29, 29] 294,912
│ └─BatchNorm2d: 2-12 [32, 256, 29, 29] 512
│ └─SiLU: 2-13 [32, 256, 29, 29] --
├─C3: 1-5 [32, 256, 29, 29] --
│ └─Conv: 2-14 [32, 128, 29, 29] --
│ │ └─Conv2d: 3-11 [32, 128, 29, 29] 32,768
│ │ └─BatchNorm2d: 3-12 [32, 128, 29, 29] 256
│ │ └─SiLU: 3-13 [32, 128, 29, 29] --
│ └─Sequential: 2-15 [32, 128, 29, 29] --
│ │ └─Bottleneck: 3-14 [32, 128, 29, 29] 164,352
│ └─Conv: 2-16 [32, 128, 29, 29] --
│ │ └─Conv2d: 3-15 [32, 128, 29, 29] 32,768
│ │ └─BatchNorm2d: 3-16 [32, 128, 29, 29] 256
│ │ └─SiLU: 3-17 [32, 128, 29, 29] --
│ └─Conv: 2-17 [32, 256, 29, 29] --
│ │ └─Conv2d: 3-18 [32, 256, 29, 29] 65,536
│ │ └─BatchNorm2d: 3-19 [32, 256, 29, 29] 512
│ │ └─SiLU: 3-20 [32, 256, 29, 29] --
├─Conv: 1-6 [32, 512, 15, 15] --
│ └─Conv2d: 2-18 [32, 512, 15, 15] 1,179,648
│ └─BatchNorm2d: 2-19 [32, 512, 15, 15] 1,024
│ └─SiLU: 2-20 [32, 512, 15, 15] --
├─C3: 1-7 [32, 512, 15, 15] --
│ └─Conv: 2-21 [32, 256, 15, 15] --
│ │ └─Conv2d: 3-21 [32, 256, 15, 15] 131,072
│ │ └─BatchNorm2d: 3-22 [32, 256, 15, 15] 512
│ │ └─SiLU: 3-23 [32, 256, 15, 15] --
│ └─Sequential: 2-22 [32, 256, 15, 15] --
│ │ └─Bottleneck: 3-24 [32, 256, 15, 15] 656,384
│ └─Conv: 2-23 [32, 256, 15, 15] --
│ │ └─Conv2d: 3-25 [32, 256, 15, 15] 131,072
│ │ └─BatchNorm2d: 3-26 [32, 256, 15, 15] 512
│ │ └─SiLU: 3-27 [32, 256, 15, 15] --
│ └─Conv: 2-24 [32, 512, 15, 15] --
│ │ └─Conv2d: 3-28 [32, 512, 15, 15] 262,144
│ │ └─BatchNorm2d: 3-29 [32, 512, 15, 15] 1,024
│ │ └─SiLU: 3-30 [32, 512, 15, 15] --
├─Conv: 1-8 [32, 1024, 8, 8] --
│ └─Conv2d: 2-25 [32, 1024, 8, 8] 4,718,592
│ └─BatchNorm2d: 2-26 [32, 1024, 8, 8] 2,048
│ └─SiLU: 2-27 [32, 1024, 8, 8] --
├─C3: 1-9 [32, 1024, 8, 8] --
│ └─Conv: 2-28 [32, 512, 8, 8] --
│ │ └─Conv2d: 3-31 [32, 512, 8, 8] 524,288
│ │ └─BatchNorm2d: 3-32 [32, 512, 8, 8] 1,024
│ │ └─SiLU: 3-33 [32, 512, 8, 8] --
│ └─Sequential: 2-29 [32, 512, 8, 8] --
│ │ └─Bottleneck: 3-34 [32, 512, 8, 8] 2,623,488
│ └─Conv: 2-30 [32, 512, 8, 8] --
│ │ └─Conv2d: 3-35 [32, 512, 8, 8] 524,288
│ │ └─BatchNorm2d: 3-36 [32, 512, 8, 8] 1,024
│ │ └─SiLU: 3-37 [32, 512, 8, 8] --
│ └─Conv: 2-31 [32, 1024, 8, 8] --
│ │ └─Conv2d: 3-38 [32, 1024, 8, 8] 1,048,576
│ │ └─BatchNorm2d: 3-39 [32, 1024, 8, 8] 2,048
│ │ └─SiLU: 3-40 [32, 1024, 8, 8] --
├─SPPF: 1-10 [32, 1024, 8, 8] --
│ └─Conv: 2-32 [32, 512, 8, 8] --
│ │ └─Conv2d: 3-41 [32, 512, 8, 8] 524,288
│ │ └─BatchNorm2d: 3-42 [32, 512, 8, 8] 1,024
│ │ └─SiLU: 3-43 [32, 512, 8, 8] --
│ └─MaxPool2d: 2-33 [32, 512, 8, 8] --
│ └─MaxPool2d: 2-34 [32, 512, 8, 8] --
│ └─MaxPool2d: 2-35 [32, 512, 8, 8] --
│ └─Conv: 2-36 [32, 1024, 8, 8] --
│ │ └─Conv2d: 3-44 [32, 1024, 8, 8] 2,097,152
│ │ └─BatchNorm2d: 3-45 [32, 1024, 8, 8] 2,048
│ │ └─SiLU: 3-46 [32, 1024, 8, 8] --
├─Sequential: 1-11 [32, 4] --
│ └─Linear: 2-37 [32, 100] 6,553,700
│ └─ReLU: 2-38 [32, 100] --
│ └─Linear: 2-39 [32, 4] 404
===============================================================================================
Total params: 21,729,592
Trainable params: 21,729,592
Non-trainable params: 0
Total mult-adds (Units.GIGABYTES): 73.80
===============================================================================================
Input size (MB): 19.27
Forward/backward pass size (MB): 2131.55
Params size (MB): 86.92
Estimated Total Size (MB): 2237.74
===============================================================================================
model = model.to(device)
三、 训练模型
1. 编写训练函数
# 训练循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset) # 训练集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
train_loss, train_acc = 0, 0 # 初始化训练损失和正确率
for X, y in dataloader: # 获取图片及其标签
X, y = X.to(device), y.to(device)
# 计算预测误差
pred = model(X) # 网络输出
loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
# 反向传播
optimizer.zero_grad() # grad属性归零
loss.backward() # 反向传播
optimizer.step() # 每一步自动更新
# 记录acc与loss
train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
train_loss += loss.item()
train_acc /= size
train_loss /= num_batches
return train_acc, train_loss
2. 编写测试函数
测试函数和训练函数大致相同，但是由于不进行梯度下降对网络权重进行更新，所以不需要传入优化器
def test (dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
test_loss, test_acc = 0, 0
# 当不进行训练时，停止梯度更新，节省计算内存消耗
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
# 计算loss
target_pred = model(imgs)
loss = loss_fn(target_pred, target)
test_loss += loss.item()
test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()
test_acc /= size
test_loss /= num_batches
return test_acc, test_loss
3. 正式训练
model.train()、model.eval()训练营往期文章中有详细的介绍。
📌如果将优化器换成 SGD 会发生什么呢？请自行探索接下来发生的诡异事件的原因。
import copy
optimizer = torch.optim.Adam(model.parameters(), lr= 1e-4)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
epochs = 60
train_loss = []
train_acc = []
test_loss = []
test_acc = []
best_acc = 0 # 设置一个最佳准确率，作为最佳模型的判别指标
for epoch in range(epochs):
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
# 保存最佳模型到 best_model
if epoch_test_acc > best_acc:
best_acc = epoch_test_acc
best_model = copy.deepcopy(model)
train_acc.append(epoch_train_acc)
train_loss.append(epoch_train_loss)
test_acc.append(epoch_test_acc)
test_loss.append(epoch_test_loss)
# 获取当前的学习率
lr = optimizer.state_dict()['param_groups'][0]['lr']
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss,
epoch_test_acc*100, epoch_test_loss, lr))
# 保存最佳模型到文件中
PATH = './best_model.pth' # 保存的参数文件名
torch.save(model.state_dict(), PATH)
print('Done')
Epoch: 1, Train_acc:54.8%, Train_loss:1.173, Test_acc:70.2%, Test_loss:0.717, Lr:1.00E-04
Epoch: 2, Train_acc:66.3%, Train_loss:0.832, Test_acc:80.4%, Test_loss:0.467, Lr:1.00E-04
Epoch: 3, Train_acc:75.0%, Train_loss:0.647, Test_acc:80.4%, Test_loss:0.515, Lr:1.00E-04
Epoch: 4, Train_acc:79.1%, Train_loss:0.562, Test_acc:85.8%, Test_loss:0.447, Lr:1.00E-04
Epoch: 5, Train_acc:84.6%, Train_loss:0.428, Test_acc:74.2%, Test_loss:0.938, Lr:1.00E-04
Epoch: 6, Train_acc:84.2%, Train_loss:0.433, Test_acc:81.8%, Test_loss:0.485, Lr:1.00E-04
Epoch: 7, Train_acc:86.2%, Train_loss:0.359, Test_acc:85.3%, Test_loss:0.408, Lr:1.00E-04
Epoch: 8, Train_acc:88.4%, Train_loss:0.331, Test_acc:93.3%, Test_loss:0.274, Lr:1.00E-04
Epoch: 9, Train_acc:89.0%, Train_loss:0.290, Test_acc:88.9%, Test_loss:0.393, Lr:1.00E-04
Epoch:10, Train_acc:89.7%, Train_loss:0.282, Test_acc:92.0%, Test_loss:0.236, Lr:1.00E-04
Epoch:11, Train_acc:92.2%, Train_loss:0.201, Test_acc:86.7%, Test_loss:0.555, Lr:1.00E-04
Epoch:12, Train_acc:92.4%, Train_loss:0.208, Test_acc:91.1%, Test_loss:0.288, Lr:1.00E-04
Epoch:13, Train_acc:92.8%, Train_loss:0.192, Test_acc:89.3%, Test_loss:0.366, Lr:1.00E-04
Epoch:14, Train_acc:94.9%, Train_loss:0.156, Test_acc:92.0%, Test_loss:0.323, Lr:1.00E-04
Epoch:15, Train_acc:92.9%, Train_loss:0.209, Test_acc:85.3%, Test_loss:0.418, Lr:1.00E-04
Epoch:16, Train_acc:93.8%, Train_loss:0.162, Test_acc:92.4%, Test_loss:0.248, Lr:1.00E-04
Epoch:17, Train_acc:95.4%, Train_loss:0.128, Test_acc:94.7%, Test_loss:0.206, Lr:1.00E-04
Epoch:18, Train_acc:96.7%, Train_loss:0.105, Test_acc:92.9%, Test_loss:0.216, Lr:1.00E-04
Epoch:19, Train_acc:97.0%, Train_loss:0.077, Test_acc:92.9%, Test_loss:0.213, Lr:1.00E-04
Epoch:20, Train_acc:98.4%, Train_loss:0.048, Test_acc:95.1%, Test_loss:0.240, Lr:1.00E-04
Epoch:21, Train_acc:96.0%, Train_loss:0.127, Test_acc:87.1%, Test_loss:0.601, Lr:1.00E-04
Epoch:22, Train_acc:96.6%, Train_loss:0.096, Test_acc:87.6%, Test_loss:0.436, Lr:1.00E-04
Epoch:23, Train_acc:97.4%, Train_loss:0.071, Test_acc:93.8%, Test_loss:0.214, Lr:1.00E-04
Epoch:24, Train_acc:98.3%, Train_loss:0.047, Test_acc:91.6%, Test_loss:0.260, Lr:1.00E-04
Epoch:25, Train_acc:97.8%, Train_loss:0.067, Test_acc:91.6%, Test_loss:0.312, Lr:1.00E-04
Epoch:26, Train_acc:96.4%, Train_loss:0.078, Test_acc:92.9%, Test_loss:0.271, Lr:1.00E-04
Epoch:27, Train_acc:96.6%, Train_loss:0.112, Test_acc:89.8%, Test_loss:0.292, Lr:1.00E-04
Epoch:28, Train_acc:97.0%, Train_loss:0.094, Test_acc:93.3%, Test_loss:0.284, Lr:1.00E-04
Epoch:29, Train_acc:96.2%, Train_loss:0.108, Test_acc:90.2%, Test_loss:0.320, Lr:1.00E-04
Epoch:30, Train_acc:96.3%, Train_loss:0.097, Test_acc:92.0%, Test_loss:0.283, Lr:1.00E-04
Epoch:31, Train_acc:99.2%, Train_loss:0.022, Test_acc:93.8%, Test_loss:0.230, Lr:1.00E-04
Epoch:32, Train_acc:99.6%, Train_loss:0.017, Test_acc:91.6%, Test_loss:0.247, Lr:1.00E-04
Epoch:33, Train_acc:98.3%, Train_loss:0.065, Test_acc:91.1%, Test_loss:0.371, Lr:1.00E-04
Epoch:34, Train_acc:96.8%, Train_loss:0.105, Test_acc:90.7%, Test_loss:0.395, Lr:1.00E-04
Epoch:35, Train_acc:98.1%, Train_loss:0.062, Test_acc:92.4%, Test_loss:0.302, Lr:1.00E-04
Epoch:36, Train_acc:96.8%, Train_loss:0.096, Test_acc:89.8%, Test_loss:0.542, Lr:1.00E-04
Epoch:37, Train_acc:97.2%, Train_loss:0.078, Test_acc:93.8%, Test_loss:0.355, Lr:1.00E-04
Epoch:38, Train_acc:99.7%, Train_loss:0.015, Test_acc:93.3%, Test_loss:0.296, Lr:1.00E-04
Epoch:39, Train_acc:100.0%, Train_loss:0.005, Test_acc:94.7%, Test_loss:0.276, Lr:1.00E-04
Epoch:40, Train_acc:99.6%, Train_loss:0.007, Test_acc:90.2%, Test_loss:0.387, Lr:1.00E-04
Epoch:41, Train_acc:98.6%, Train_loss:0.044, Test_acc:87.6%, Test_loss:0.474, Lr:1.00E-04
Epoch:42, Train_acc:99.1%, Train_loss:0.028, Test_acc:83.6%, Test_loss:0.987, Lr:1.00E-04
Epoch:43, Train_acc:96.6%, Train_loss:0.134, Test_acc:86.2%, Test_loss:1.002, Lr:1.00E-04
Epoch:44, Train_acc:97.3%, Train_loss:0.078, Test_acc:90.2%, Test_loss:0.525, Lr:1.00E-04
Epoch:45, Train_acc:98.6%, Train_loss:0.031, Test_acc:88.0%, Test_loss:0.623, Lr:1.00E-04
Epoch:46, Train_acc:99.4%, Train_loss:0.016, Test_acc:93.8%, Test_loss:0.307, Lr:1.00E-04
Epoch:47, Train_acc:100.0%, Train_loss:0.003, Test_acc:93.8%, Test_loss:0.289, Lr:1.00E-04
Epoch:48, Train_acc:98.2%, Train_loss:0.066, Test_acc:88.4%, Test_loss:0.464, Lr:1.00E-04
Epoch:49, Train_acc:98.8%, Train_loss:0.047, Test_acc:90.2%, Test_loss:0.339, Lr:1.00E-04
Epoch:50, Train_acc:97.0%, Train_loss:0.089, Test_acc:89.3%, Test_loss:0.523, Lr:1.00E-04
Epoch:51, Train_acc:99.2%, Train_loss:0.026, Test_acc:89.3%, Test_loss:0.517, Lr:1.00E-04
Epoch:52, Train_acc:99.4%, Train_loss:0.032, Test_acc:91.1%, Test_loss:0.343, Lr:1.00E-04
Epoch:53, Train_acc:99.2%, Train_loss:0.027, Test_acc:92.9%, Test_loss:0.612, Lr:1.00E-04
Epoch:54, Train_acc:99.2%, Train_loss:0.019, Test_acc:91.6%, Test_loss:0.461, Lr:1.00E-04
Epoch:55, Train_acc:99.9%, Train_loss:0.004, Test_acc:92.4%, Test_loss:0.480, Lr:1.00E-04
Epoch:56, Train_acc:99.7%, Train_loss:0.005, Test_acc:92.4%, Test_loss:0.463, Lr:1.00E-04
Epoch:57, Train_acc:100.0%, Train_loss:0.004, Test_acc:92.0%, Test_loss:0.468, Lr:1.00E-04
Epoch:58, Train_acc:98.1%, Train_loss:0.079, Test_acc:85.8%, Test_loss:0.686, Lr:1.00E-04
Epoch:59, Train_acc:96.8%, Train_loss:0.129, Test_acc:92.0%, Test_loss:0.357, Lr:1.00E-04
Epoch:60, Train_acc:97.4%, Train_loss:0.075, Test_acc:91.1%, Test_loss:0.467, Lr:1.00E-04
Done
四、 结果可视化
1. Loss与Accuracy图
import matplotlib.pyplot as plt
#隐藏警告
import warnings
warnings.filterwarnings("ignore") #忽略警告信息
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
plt.rcParams['figure.dpi'] = 100 #分辨率
epochs_range = range(epochs)
plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
2. 模型评估
best_model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)
epoch_test_acc, epoch_test_loss
(0.9511111111111111, 0.2404080551484201)
# 查看是否与我们记录的最高准确率一致
epoch_test_acc
0.9511111111111111
五、测试感想
本以为这个是完整的yolov5, 运行完后才发现仅仅是backbone部分，还缺少neck和head.
YOLOv5中引入了大量的缩写，影响了对网络架构的理解。简单罗列如下，还需要每个模块深入了解
- CBS: Conv + BatchNorm + Silu
- C3: Concentrated-Comprehensive Convolution
a. https://sh-tsang.medium.com/reading-c3-concentrated-comprehensive-convolution-semantic-segmentation-5b6dd3fb46b2 - CSPNet: Cross Stage Partial Network
- FPN: Feature Pyramid Network
- PAN: Pyramid Attention Network
- SPPF: Spatial Pyramid Pooling - Fast (SPPF)
扫一扫
2045
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
