河北区网站建设,微企免费网站建设,应用商店网站模板,广东网站建设网#x1f368; 本文为#x1f517;365天深度学习训练营 中的学习记录博客#x1f356; 原作者#xff1a;K同学啊
前言
yolov5网络结构比较复杂#xff0c;上次我们简要介绍了yolov5网络模块#xff0c;并且复现了C3模块#xff0c;深度学习基础–yolov5网络结构简介 本文为365天深度学习训练营 中的学习记录博客 原作者K同学啊
前言
yolov5网络结构比较复杂上次我们简要介绍了yolov5网络模块并且复现了C3模块深度学习基础–yolov5网络结构简介C3模块构建;这一次我们将复现backbone模块将目标检测网络结构用到目标识别上会是怎样的效果呢这周是考试周周一到周四一直都在准备考试和去考试昨天开始又发高烧更新较慢欢迎收藏加关注本人将会持续更新。 文章目录 案例1、数据处理1、导入库2、查看数据类别3、导入数据4、数据集划分5、展示一批数据 2、模型构建3、模型训练1、构建训练集2、构建测试集3、设置超参数 4、模型正式训练5、结果显示和评估1、结果显示2、评估
案例
将backbone模块用于识别天气分类
1、数据处理
1、导入库
import torch
import torchvision
import torch.nn as nn
import torchvision.transforms as transforms
from torchvision import datasets, transformsdevice cuda if torch.cuda.is_available() else cpudevicecuda2、查看数据类别
import os, pathlib data_dir ./data/
data_dir pathlib.Path(data_dir)classnames [str(path).split(\\)[0] for path in os.listdir(data_dir)]
classnames[cloudy, rain, shine, sunrise]3、导入数据
data_transforms transforms.Compose([transforms.Resize([224, 224]),transforms.ToTensor(),transforms.Normalize( # 数据标准化处理--- 转化为 标准状态分布使模型更容易收敛mean[0.485, 0.456, 0.406], # rgb均值std[0.229, 0.224, 0.225] # rgb标准差这两个从数据集中随机抽样得到的)
])total_data datasets.ImageFolder(./data/, data_transforms)
total_dataDataset ImageFolderNumber of datapoints: 1125Root location: ./data/StandardTransform
Transform: Compose(Resize(size[224, 224], interpolationbilinear, max_sizeNone, antialiasTrue)ToTensor()Normalize(mean[0.485, 0.456, 0.406], std[0.229, 0.224, 0.225]))4、数据集划分
训练集 测试集 8 2
train_size int(len(total_data) * 0.8)
test_size len(total_data) - train_size
train_data, test_data torch.utils.data.random_split(total_data, [train_size, test_size])
print(train_size, len(train_data))
print(test_size, len(test_data))train_size 900
test_size 225# 动态加载数据集
batch_size 4train_dl torch.utils.data.DataLoader(train_data,batch_sizebatch_size,shuffleTrue
)test_dl torch.utils.data.DataLoader(test_data,batch_sizebatch_size,shuffleTrue
)# 查看数据格式
temp_data, temp_label next(iter(train_dl))print(data: , temp_data.shape)
print(data_labels: , temp_label)data: torch.Size([4, 3, 224, 224])
data_labels: tensor([3, 2, 0, 0])5、展示一批数据
这里一批次大小4
import matplotlib.pyplot as plt temp_images, temp_labels next(iter(test_dl))plt.figure(figsize(20, 10))for i in range(4):plt.subplot(5, 5, i 1)plt.imshow(temp_images[i].cpu().numpy().transpose(1, 2, 0)) # C H W) H W Cplt.title(classnames[temp_labels[i]])plt.axis(off)plt.show()2、模型构建
整体网络 C3网络参考深度学习基础–yolov5网络结构简介C3模块构建
SPPF网络模块图结构如下 import torch.nn.functional as F
import warnings # 确保导入 warnings 模块# 自动计算p(填充)
def autop(k, pNone):if p is None:p k // 2 if isinstance(k, int) else [i // 2 for i in k]return p # Conv模块搭建卷积层 标准化 激活函数class Conv(nn.Module):def __init__(self, c1, c2, k1, s1, pNone, g1, actTrue):super().__init__()groups: 1: 标准卷积c1: 深度卷积1 ~ c1: 分组卷积bias:false: 不使用偏置self.conv nn.Conv2d(c1, c2, kernel_sizek, strides, paddingautop(k, p), groupsg, biasFalse) self.bn nn.BatchNorm2d(c2)act:true: silu激活函数否则: 如果是nn.Mudule(如: nn.Relu), 则调用本身否则: Identity, 什么都不做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)))# Bottleneck模块, 用于特征提取和用于防止梯度消失、梯度爆炸问题
class Bottleneck(nn.Module):def __init__(self, c1, c2, shortcutTrue, g1, e0.5): # shortcut: 是否需要残差连接 e: 模型深度super().__init__()c_ int(c1 * 2)self.cv1 Conv(c1, c_, 1, 1)self.cv2 Conv(c_, c2, 3, 1, gg)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))# 搭建C3模块
class C3(nn.Module):def __init__(self, c1, c2, n1, shortcutTrue, g1, e0.5):super().__init__()刚开始: 卷积层2层后面: n层bottlenck后 concat后 convc_ int(c1 * e)self.cv1 Conv(c1, c_, 1, 1)self.cv2 Conv(c1, c_, 1, 1) # 用于拼接self.cv3 Conv(2 * c_, c2, 1)self.m nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e1.0) for _ in range(n))) # * 解包def forward(self, x):# 拼接按照 dim1维进行拼接故列要相同return self.cv3(torch.cat([self.m(self.cv1(x)), self.cv2(x)], dim1)) # 结合图就知道了结构# 搭建SPPF模块用于特征融合
class SPPF(nn.Module):def __init__(self, c1, c2, k5):super().__init__()c_ c1 // 2 self.cv1 Conv(c1, c_, 1, 1)self.cv2 Conv(c_ * 4, c2, 1, 1) # 模型融合这个时候模型通道扩大4倍套用池化层公式发现通过.m 通道数数不变self.m nn.MaxPool2d(kernel_sizek, stride1, paddingk // 2) # 套用卷积层、池化层公式发现输出通道不变def forward(self, x):x self.cv1(x)with warnings.catch_warnings():warnings.simplefilter(ignore)y1 self.m(x)y2 self.m(y1)y3 self.m(y2)return self.cv2(torch.cat([x, y1, y2, y3], 1)) # 结合图# 搭建backbone模块
class Yolov5_backbone(nn.Module):def __init__(self):super(Yolov5_backbone, self).__init__()# 采用常规卷积, kernel_size, stride 与 yolov5.yaml配置文件一致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_10 SPPF(1024, 1024, 5)self.classifiler nn.Sequential(nn.Linear(in_features65536, out_features100),nn.ReLU(),nn.Linear(in_features100, out_featureslen(classnames)))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_10(x)x torch.flatten(x, start_dim1)x self.classifiler(x)return x# 输出参数
model Yolov5_backbone().to(device)
modelYolov5_backbone((conv_1): Conv((conv): Conv2d(3, 64, kernel_size(3, 3), stride(2, 2), padding(2, 2), biasFalse)(bn): BatchNorm2d(64, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(conv_2): Conv((conv): Conv2d(64, 128, kernel_size(3, 3), stride(2, 2), padding(1, 1), biasFalse)(bn): BatchNorm2d(128, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(c3_3): C3((cv1): Conv((conv): Conv2d(128, 64, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(64, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(128, 64, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(64, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv3): Conv((conv): Conv2d(128, 128, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(128, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(64, 128, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(128, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(128, 64, kernel_size(3, 3), stride(1, 1), padding(1, 1), biasFalse)(bn): BatchNorm2d(64, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU()))))(conv_4): Conv((conv): Conv2d(128, 256, kernel_size(3, 3), stride(2, 2), padding(1, 1), biasFalse)(bn): BatchNorm2d(256, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(c3_5): C3((cv1): Conv((conv): Conv2d(256, 128, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(128, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(256, 128, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(128, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv3): Conv((conv): Conv2d(256, 256, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(256, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(128, 256, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(256, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(256, 128, kernel_size(3, 3), stride(1, 1), padding(1, 1), biasFalse)(bn): BatchNorm2d(128, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU()))))(conv_6): Conv((conv): Conv2d(256, 512, kernel_size(3, 3), stride(2, 2), padding(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(c3_7): C3((cv1): Conv((conv): Conv2d(512, 256, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(256, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(512, 256, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(256, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv3): Conv((conv): Conv2d(512, 512, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(256, 512, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(512, 256, kernel_size(3, 3), stride(1, 1), padding(1, 1), biasFalse)(bn): BatchNorm2d(256, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU()))))(conv_8): Conv((conv): Conv2d(512, 1024, kernel_size(3, 3), stride(2, 2), padding(1, 1), biasFalse)(bn): BatchNorm2d(1024, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(c3_9): C3((cv1): Conv((conv): Conv2d(1024, 512, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(1024, 512, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv3): Conv((conv): Conv2d(1024, 1024, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(1024, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(m): Sequential((0): Bottleneck((cv1): Conv((conv): Conv2d(512, 1024, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(1024, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(1024, 512, kernel_size(3, 3), stride(1, 1), padding(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU()))))(SPPF_10): SPPF((cv1): Conv((conv): Conv2d(1024, 512, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(512, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(cv2): Conv((conv): Conv2d(2048, 1024, kernel_size(1, 1), stride(1, 1), biasFalse)(bn): BatchNorm2d(1024, eps1e-05, momentum0.1, affineTrue, track_running_statsTrue)(act): SiLU())(m): MaxPool2d(kernel_size5, stride1, padding2, dilation1, ceil_modeFalse))(classifiler): Sequential((0): Linear(in_features65536, out_features100, biasTrue)(1): ReLU()(2): Linear(in_features100, out_features4, biasTrue))
)import torchsummary as summary
summary.summary(model, (3, 224, 224))----------------------------------------------------------------Layer (type) Output Shape Param #
Conv2d-1 [-1, 64, 113, 113] 1,728BatchNorm2d-2 [-1, 64, 113, 113] 128SiLU-3 [-1, 64, 113, 113] 0Conv-4 [-1, 64, 113, 113] 0Conv2d-5 [-1, 128, 57, 57] 73,728BatchNorm2d-6 [-1, 128, 57, 57] 256SiLU-7 [-1, 128, 57, 57] 0Conv-8 [-1, 128, 57, 57] 0Conv2d-9 [-1, 64, 57, 57] 8,192BatchNorm2d-10 [-1, 64, 57, 57] 128SiLU-11 [-1, 64, 57, 57] 0Conv-12 [-1, 64, 57, 57] 0Conv2d-13 [-1, 128, 57, 57] 8,192BatchNorm2d-14 [-1, 128, 57, 57] 256SiLU-15 [-1, 128, 57, 57] 0Conv-16 [-1, 128, 57, 57] 0Conv2d-17 [-1, 64, 57, 57] 73,728BatchNorm2d-18 [-1, 64, 57, 57] 128SiLU-19 [-1, 64, 57, 57] 0Conv-20 [-1, 64, 57, 57] 0Bottleneck-21 [-1, 64, 57, 57] 0Conv2d-22 [-1, 64, 57, 57] 8,192BatchNorm2d-23 [-1, 64, 57, 57] 128SiLU-24 [-1, 64, 57, 57] 0Conv-25 [-1, 64, 57, 57] 0Conv2d-26 [-1, 128, 57, 57] 16,384BatchNorm2d-27 [-1, 128, 57, 57] 256SiLU-28 [-1, 128, 57, 57] 0Conv-29 [-1, 128, 57, 57] 0C3-30 [-1, 128, 57, 57] 0Conv2d-31 [-1, 256, 29, 29] 294,912BatchNorm2d-32 [-1, 256, 29, 29] 512SiLU-33 [-1, 256, 29, 29] 0Conv-34 [-1, 256, 29, 29] 0Conv2d-35 [-1, 128, 29, 29] 32,768BatchNorm2d-36 [-1, 128, 29, 29] 256SiLU-37 [-1, 128, 29, 29] 0Conv-38 [-1, 128, 29, 29] 0Conv2d-39 [-1, 256, 29, 29] 32,768BatchNorm2d-40 [-1, 256, 29, 29] 512SiLU-41 [-1, 256, 29, 29] 0Conv-42 [-1, 256, 29, 29] 0Conv2d-43 [-1, 128, 29, 29] 294,912BatchNorm2d-44 [-1, 128, 29, 29] 256SiLU-45 [-1, 128, 29, 29] 0Conv-46 [-1, 128, 29, 29] 0Bottleneck-47 [-1, 128, 29, 29] 0Conv2d-48 [-1, 128, 29, 29] 32,768BatchNorm2d-49 [-1, 128, 29, 29] 256SiLU-50 [-1, 128, 29, 29] 0Conv-51 [-1, 128, 29, 29] 0Conv2d-52 [-1, 256, 29, 29] 65,536BatchNorm2d-53 [-1, 256, 29, 29] 512SiLU-54 [-1, 256, 29, 29] 0Conv-55 [-1, 256, 29, 29] 0C3-56 [-1, 256, 29, 29] 0Conv2d-57 [-1, 512, 15, 15] 1,179,648BatchNorm2d-58 [-1, 512, 15, 15] 1,024SiLU-59 [-1, 512, 15, 15] 0Conv-60 [-1, 512, 15, 15] 0Conv2d-61 [-1, 256, 15, 15] 131,072BatchNorm2d-62 [-1, 256, 15, 15] 512SiLU-63 [-1, 256, 15, 15] 0Conv-64 [-1, 256, 15, 15] 0Conv2d-65 [-1, 512, 15, 15] 131,072BatchNorm2d-66 [-1, 512, 15, 15] 1,024SiLU-67 [-1, 512, 15, 15] 0Conv-68 [-1, 512, 15, 15] 0Conv2d-69 [-1, 256, 15, 15] 1,179,648BatchNorm2d-70 [-1, 256, 15, 15] 512SiLU-71 [-1, 256, 15, 15] 0Conv-72 [-1, 256, 15, 15] 0Bottleneck-73 [-1, 256, 15, 15] 0Conv2d-74 [-1, 256, 15, 15] 131,072BatchNorm2d-75 [-1, 256, 15, 15] 512SiLU-76 [-1, 256, 15, 15] 0Conv-77 [-1, 256, 15, 15] 0Conv2d-78 [-1, 512, 15, 15] 262,144BatchNorm2d-79 [-1, 512, 15, 15] 1,024SiLU-80 [-1, 512, 15, 15] 0Conv-81 [-1, 512, 15, 15] 0C3-82 [-1, 512, 15, 15] 0Conv2d-83 [-1, 1024, 8, 8] 4,718,592BatchNorm2d-84 [-1, 1024, 8, 8] 2,048SiLU-85 [-1, 1024, 8, 8] 0Conv-86 [-1, 1024, 8, 8] 0Conv2d-87 [-1, 512, 8, 8] 524,288BatchNorm2d-88 [-1, 512, 8, 8] 1,024SiLU-89 [-1, 512, 8, 8] 0Conv-90 [-1, 512, 8, 8] 0Conv2d-91 [-1, 1024, 8, 8] 524,288BatchNorm2d-92 [-1, 1024, 8, 8] 2,048SiLU-93 [-1, 1024, 8, 8] 0Conv-94 [-1, 1024, 8, 8] 0Conv2d-95 [-1, 512, 8, 8] 4,718,592BatchNorm2d-96 [-1, 512, 8, 8] 1,024SiLU-97 [-1, 512, 8, 8] 0Conv-98 [-1, 512, 8, 8] 0Bottleneck-99 [-1, 512, 8, 8] 0Conv2d-100 [-1, 512, 8, 8] 524,288BatchNorm2d-101 [-1, 512, 8, 8] 1,024SiLU-102 [-1, 512, 8, 8] 0Conv-103 [-1, 512, 8, 8] 0Conv2d-104 [-1, 1024, 8, 8] 1,048,576BatchNorm2d-105 [-1, 1024, 8, 8] 2,048SiLU-106 [-1, 1024, 8, 8] 0Conv-107 [-1, 1024, 8, 8] 0C3-108 [-1, 1024, 8, 8] 0Conv2d-109 [-1, 512, 8, 8] 524,288BatchNorm2d-110 [-1, 512, 8, 8] 1,024SiLU-111 [-1, 512, 8, 8] 0Conv-112 [-1, 512, 8, 8] 0MaxPool2d-113 [-1, 512, 8, 8] 0MaxPool2d-114 [-1, 512, 8, 8] 0MaxPool2d-115 [-1, 512, 8, 8] 0Conv2d-116 [-1, 1024, 8, 8] 2,097,152BatchNorm2d-117 [-1, 1024, 8, 8] 2,048SiLU-118 [-1, 1024, 8, 8] 0Conv-119 [-1, 1024, 8, 8] 0SPPF-120 [-1, 1024, 8, 8] 0Linear-121 [-1, 100] 6,553,700ReLU-122 [-1, 100] 0Linear-123 [-1, 4] 404Total params: 25,213,112
Trainable params: 25,213,112
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 149.98
Params size (MB): 96.18
Estimated Total Size (MB): 246.74
----------------------------------------------------------------3、模型训练
1、构建训练集
def train(dataloader, model, loss_fn, optimizer):size len(dataloader.dataset) # 总数目num_batch len(dataloader) # 批次数目train_acc, train_loss 0, 0for X, y in dataloader:X, y X.to(device), y.to(device) predict model(X)loss loss_fn(predict, y)# 梯度清0、求导、重新设置参数optimizer.zero_grad()loss.backward()optimizer.step()train_acc (predict.argmax(1) y).type(torch.float).sum().item()train_loss loss.item()train_acc / sizetrain_loss / num_batchreturn train_acc, train_loss2、构建测试集
def test(dataloader, model, loss_fn):size len(dataloader.dataset)num_batch len(dataloader)test_acc, test_loss 0, 0with torch.no_grad():for X, y in dataloader:X, y X.to(device), y.to(device)predict model(X)loss loss_fn(predict, y)test_acc (predict.argmax(1) y).type(torch.float).sum().item()test_loss loss.item()test_acc / size test_loss / num_batchreturn test_acc, test_loss 3、设置超参数
learn_rate 1e-4
optimizer torch.optim.Adam(model.parameters(), lrlearn_rate)
loss_fn nn.CrossEntropyLoss()4、模型正式训练
import copy epochs 60train_acc, train_loss, test_acc, test_loss [], [], [], []best_acc 0for 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_modelif epoch_test_acc best_acc:best_acc epoch_test_accbest_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(epoch1, 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:59.0%, Train_loss:1.106, Test_acc:71.1%, Test_loss:0.740, Lr:1.00E-04
Epoch: 2, Train_acc:70.0%, Train_loss:0.776, Test_acc:83.1%, Test_loss:0.468, Lr:1.00E-04
Epoch: 3, Train_acc:75.2%, Train_loss:0.661, Test_acc:84.0%, Test_loss:0.461, Lr:1.00E-04
Epoch: 4, Train_acc:77.4%, Train_loss:0.605, Test_acc:88.4%, Test_loss:0.396, Lr:1.00E-04
Epoch: 5, Train_acc:80.3%, Train_loss:0.529, Test_acc:82.7%, Test_loss:0.415, Lr:1.00E-04
Epoch: 6, Train_acc:83.8%, Train_loss:0.422, Test_acc:83.6%, Test_loss:0.416, Lr:1.00E-04
Epoch: 7, Train_acc:85.8%, Train_loss:0.423, Test_acc:87.1%, Test_loss:0.343, Lr:1.00E-04
Epoch: 8, Train_acc:85.6%, Train_loss:0.393, Test_acc:87.6%, Test_loss:0.306, Lr:1.00E-04
Epoch: 9, Train_acc:86.3%, Train_loss:0.354, Test_acc:89.3%, Test_loss:0.338, Lr:1.00E-04
Epoch:10, Train_acc:86.6%, Train_loss:0.340, Test_acc:92.9%, Test_loss:0.276, Lr:1.00E-04
Epoch:11, Train_acc:88.8%, Train_loss:0.317, Test_acc:90.2%, Test_loss:0.290, Lr:1.00E-04
Epoch:12, Train_acc:87.9%, Train_loss:0.327, Test_acc:88.0%, Test_loss:0.338, Lr:1.00E-04
Epoch:13, Train_acc:89.2%, Train_loss:0.315, Test_acc:92.0%, Test_loss:0.337, Lr:1.00E-04
Epoch:14, Train_acc:91.1%, Train_loss:0.230, Test_acc:92.0%, Test_loss:0.369, Lr:1.00E-04
Epoch:15, Train_acc:93.3%, Train_loss:0.182, Test_acc:89.8%, Test_loss:0.278, Lr:1.00E-04
Epoch:16, Train_acc:91.6%, Train_loss:0.229, Test_acc:90.2%, Test_loss:0.290, Lr:1.00E-04
Epoch:17, Train_acc:90.9%, Train_loss:0.230, Test_acc:91.6%, Test_loss:0.272, Lr:1.00E-04
Epoch:18, Train_acc:93.9%, Train_loss:0.152, Test_acc:92.0%, Test_loss:0.280, Lr:1.00E-04
Epoch:19, Train_acc:94.7%, Train_loss:0.159, Test_acc:92.0%, Test_loss:0.262, Lr:1.00E-04
Epoch:20, Train_acc:95.9%, Train_loss:0.124, Test_acc:91.1%, Test_loss:0.260, Lr:1.00E-04
Epoch:21, Train_acc:95.7%, Train_loss:0.102, Test_acc:88.9%, Test_loss:0.342, Lr:1.00E-04
Epoch:22, Train_acc:95.9%, Train_loss:0.113, Test_acc:92.4%, Test_loss:0.275, Lr:1.00E-04
Epoch:23, Train_acc:96.1%, Train_loss:0.130, Test_acc:92.9%, Test_loss:0.308, Lr:1.00E-04
Epoch:24, Train_acc:94.8%, Train_loss:0.161, Test_acc:86.7%, Test_loss:0.456, Lr:1.00E-04
Epoch:25, Train_acc:95.2%, Train_loss:0.139, Test_acc:89.3%, Test_loss:0.428, Lr:1.00E-04
Epoch:26, Train_acc:96.0%, Train_loss:0.103, Test_acc:92.9%, Test_loss:0.313, Lr:1.00E-04
Epoch:27, Train_acc:96.0%, Train_loss:0.098, Test_acc:88.9%, Test_loss:0.520, Lr:1.00E-04
Epoch:28, Train_acc:97.2%, Train_loss:0.079, Test_acc:91.1%, Test_loss:0.404, Lr:1.00E-04
Epoch:29, Train_acc:98.6%, Train_loss:0.037, Test_acc:92.0%, Test_loss:0.270, Lr:1.00E-04
Epoch:30, Train_acc:98.8%, Train_loss:0.033, Test_acc:88.4%, Test_loss:0.520, Lr:1.00E-04
Epoch:31, Train_acc:95.6%, Train_loss:0.139, Test_acc:91.6%, Test_loss:0.370, Lr:1.00E-04
Epoch:32, Train_acc:96.7%, Train_loss:0.116, Test_acc:89.3%, Test_loss:0.376, Lr:1.00E-04
Epoch:33, Train_acc:96.4%, Train_loss:0.102, Test_acc:91.6%, Test_loss:0.342, Lr:1.00E-04
Epoch:34, Train_acc:98.6%, Train_loss:0.049, Test_acc:87.1%, Test_loss:0.417, Lr:1.00E-04
Epoch:35, Train_acc:97.9%, Train_loss:0.068, Test_acc:90.7%, Test_loss:0.423, Lr:1.00E-04
Epoch:36, Train_acc:98.3%, Train_loss:0.048, Test_acc:89.3%, Test_loss:0.492, Lr:1.00E-04
Epoch:37, Train_acc:98.0%, Train_loss:0.054, Test_acc:91.1%, Test_loss:0.355, Lr:1.00E-04
Epoch:38, Train_acc:98.6%, Train_loss:0.060, Test_acc:92.4%, Test_loss:0.402, Lr:1.00E-04
Epoch:39, Train_acc:97.9%, Train_loss:0.065, Test_acc:86.7%, Test_loss:0.498, Lr:1.00E-04
Epoch:40, Train_acc:98.0%, Train_loss:0.055, Test_acc:88.4%, Test_loss:0.514, Lr:1.00E-04
Epoch:41, Train_acc:99.1%, Train_loss:0.029, Test_acc:90.7%, Test_loss:0.381, Lr:1.00E-04
Epoch:42, Train_acc:98.0%, Train_loss:0.069, Test_acc:92.4%, Test_loss:0.377, Lr:1.00E-04
Epoch:43, Train_acc:99.4%, Train_loss:0.021, Test_acc:90.2%, Test_loss:0.403, Lr:1.00E-04
Epoch:44, Train_acc:98.0%, Train_loss:0.055, Test_acc:85.3%, Test_loss:0.686, Lr:1.00E-04
Epoch:45, Train_acc:98.0%, Train_loss:0.074, Test_acc:91.1%, Test_loss:0.321, Lr:1.00E-04
Epoch:46, Train_acc:98.6%, Train_loss:0.038, Test_acc:91.6%, Test_loss:0.426, Lr:1.00E-04
Epoch:47, Train_acc:97.4%, Train_loss:0.075, Test_acc:87.1%, Test_loss:0.604, Lr:1.00E-04
Epoch:48, Train_acc:99.6%, Train_loss:0.027, Test_acc:91.6%, Test_loss:0.379, Lr:1.00E-04
Epoch:49, Train_acc:99.8%, Train_loss:0.007, Test_acc:92.4%, Test_loss:0.381, Lr:1.00E-04
Epoch:50, Train_acc:100.0%, Train_loss:0.007, Test_acc:92.9%, Test_loss:0.361, Lr:1.00E-04
Epoch:51, Train_acc:99.8%, Train_loss:0.018, Test_acc:90.7%, Test_loss:0.446, Lr:1.00E-04
Epoch:52, Train_acc:99.0%, Train_loss:0.032, Test_acc:89.8%, Test_loss:0.588, Lr:1.00E-04
Epoch:53, Train_acc:97.7%, Train_loss:0.060, Test_acc:90.7%, Test_loss:0.456, Lr:1.00E-04
Epoch:54, Train_acc:97.7%, Train_loss:0.059, Test_acc:89.8%, Test_loss:0.506, Lr:1.00E-04
Epoch:55, Train_acc:98.6%, Train_loss:0.046, Test_acc:90.7%, Test_loss:0.350, Lr:1.00E-04
Epoch:56, Train_acc:99.7%, Train_loss:0.010, Test_acc:91.6%, Test_loss:0.349, Lr:1.00E-04
Epoch:57, Train_acc:99.6%, Train_loss:0.012, Test_acc:91.6%, Test_loss:0.369, Lr:1.00E-04
Epoch:58, Train_acc:98.9%, Train_loss:0.053, Test_acc:88.9%, Test_loss:0.666, Lr:1.00E-04
Epoch:59, Train_acc:98.2%, Train_loss:0.054, Test_acc:87.1%, Test_loss:0.509, Lr:1.00E-04
Epoch:60, Train_acc:98.7%, Train_loss:0.037, Test_acc:90.2%, Test_loss:0.513, Lr:1.00E-04
Done5、结果显示和评估
1、结果显示
import matplotlib.pyplot as plt #隐藏警告和显示中文
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x range(epochs)
# 创建画板
plt.figure(figsize(12, 3))
# 子图一
plt.subplot(1, 2, 1)
plt.plot(x, train_acc, labelTrain Accurary)
plt.plot(x, test_acc, labelTest Accurary)
plt.legend(loclower right)
plt.title(Train and test Accurary)
# 子图二
plt.subplot(1, 2, 2)
plt.plot(x, train_loss, labelTrain loss)
plt.plot(x, test_loss, labelTest loss)
plt.legend(locupper right)
plt.title(Train and test Loss)plt.show()解释
总体效果还是不错的损失率低于1但是测试集的损失率有点小小不稳定准确率刚开始出现了欠拟合的现象但是后面好了训练准确率稳定在100%附件(98%、99%等)测试集稳定在90%附件整体yolov5这个用于目标检测的网络用语目标识别也是有不错的效果。
2、评估
best_model.load_state_dict(torch.load(PATH, map_locationdevice))
epoch_test_acc, epoch_test_loss test(test_dl, best_model, loss_fn)epoch_test_acc, epoch_test_loss(0.9022222222222223, 0.5129974257313852)准确率在0.9左右效果良好且损失率为0.5低于1.0。
