（7）catanddog_用mobilenetv3迁移学习分类和测试

chencaw

已于 2022-11-04 14:50:54 修改

阅读量149

点赞数

分类专栏： torch 文章标签：迁移学习分类深度学习

于 2022-10-12 15:30:57 首次发布

本文链接：https://blog.csdn.net/chencaw/article/details/127283569

版权

1、网络的介绍参照下面

MobileNet系列(4):MobileNetv3网络详解_@BangBang的博客-CSDN博客_mobilenetv3

和下面的文字

https://www.toutiao.com/article/6972051587388785184/?app=news_article&timestamp=1665491325&use_new_style=1&req_id=202210112028440101320380301D113152&group_id=6972051587388785184&tt_from=mobile_qq&utm_source=mobile_qq&utm_medium=toutiao_android&utm_campaign=client_share&share_token=b18046b3-f26e-473e-b539-1fc65dfd400d&source=m_redirect

TorchVision v0.9 中新增了一系列移动端友好的模型，可用于处理分类、目标检测、语义分割等任务。

2、研究mobilenetv3最后的输出层

3、修改输出层的方法

3.1 输出层修改实现方法一

    chen_set_mode_out = nn.Sequential(
                            nn.Flatten(),
                            nn.Linear(576,1024),
                            nn.Hardswish(inplace=True),
                            nn.Dropout(p=0.2, inplace=True),
                            nn.Linear(1024, 1000),
                            nn.Linear(1000, 2)
                          ).to(device)


    # small = mobilenet_v3_small(pretrained=True)
    trained_model = mobilenet_v3_small(weights=models.MobileNet_V3_Small_Weights.DEFAULT)
    #注意默认是1000分类输出，[256,1000]
    
    #（1）实现方法一：

    model = nn.Sequential(*list(trained_model.children())[:-1], #[b, 512, 1, 1]
                        #   Flatten(), # [b, 512, 1, 1] => [b, 512]
                        #   nn.Flatten(),
                        #   nn.Linear(256, 2)
                        chen_set_mode_out
                          ).to(device)

3.2 输出层修改实现方法二

    #（2）实现方法二
    print("陈打印最后的分类模块---------------------")
    print(trained_model.classifier._modules['0'])
    print(trained_model.classifier._modules['1'])
    print(trained_model.classifier._modules['2'])
    print(trained_model.classifier._modules['3'])
    trained_model.classifier._modules['3'] = nn.Linear(1024, 2)
    model = trained_model

4、代码实现

4.1方法一的代码

import  torch
from    torch import optim, nn
# import  visdom
# from tensorboardX import SummaryWriter  #(1)引入tensorboardX
from torch.utils.tensorboard import SummaryWriter
 
# import  torchvision
from    torch.utils.data import DataLoader
from    torchvision import transforms, datasets, models
# from    utils import Flatten
 
# from    pokemon import Pokemon
# from    resnet import ResNet18
# from    torchvision.models import resnet18
from    torchvision.models import  mobilenet_v3_small
from    PIL import Image
from    tqdm import tqdm
from    torchinfo import summary
import  os
 
 
# batchsz = 32
batch_size = 256
lr = 1e-3
epochs = 10
img_resize = 224
 
# print("cuda:") 
# print(torch.cuda.is_available())
# print(torch.cuda.device_count())
# print(torch.cuda.current_device()) 
# print(torch.cuda.get_device_name(0)) 
 
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# device = torch.device('cpu')
torch.manual_seed(1234)
 
# tf = transforms.Compose([
#                      transforms.Resize((224,224)),
#                      transforms.ToTensor(),
#      ])
#输入应该是PIL.Image类型
tf = transforms.Compose([
    #匿名函数
    # lambda x:Image.open(x).convert('RGB'), # string path= > image data
    transforms.Resize((int(img_resize*1.25), int(img_resize*1.25))),
    transforms.RandomRotation(15),
    transforms.CenterCrop(img_resize),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406],
                            std=[0.229, 0.224, 0.225])
])
# db = torchvision.datasets.ImageFolder(root=&#