10
PyTorch 迁移学习之模型结构修改和参数微调
source link: https://xujinzh.github.io/2023/06/05/pytorch-fine-tuning-model/
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.
PyTorch 迁移学习之模型结构修改和参数微调
发表于2023-06-05|更新于2023-06-05|technologypython
字数总计:7.4k|阅读时长:42分钟|阅读量:3
PyTorch 迁移学习实例。使用 resnet 模型框架在 cifar10 数据集上进行模型迁移学习。
导入依赖包
import copy
import os
from datetime import datetime
import matplotlib.pyplot as plt
import numpy as np
import torch
import torch.optim as optim
import visdom
from torch import nn
from torch.utils.data import DataLoader
from torchvision import datasets, models, transforms
torch.manual_seed(33)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
数据读取与预处理操作
数据结构数如下：
$ tree -L 2 /disk1/datasets/cifar10/cifar10
/disk1/datasets/cifar10/cifar10
├── test
│ ├── airplane
│ ├── automobile
│ ├── bird
│ ├── cat
│ ├── deer
│ ├── dog
│ ├── frog
│ ├── horse
│ ├── ship
│ └── truck
└── train
├── airplane
├── automobile
├── bird
├── cat
├── deer
├── dog
├── frog
├── horse
├── ship
└── truck
22 directories, 0 files
其中每个文件夹（如：/disk1/datasets/cifar10/cifar10/train/airplane）里面都是图片，类似：
$ ls -l /disk1/datasets/cifar10/cifar10/train/airplane | head -n 5
total 20000
-rwxrwxr-x 1 jinzhongxu jinzhongxu 2063 3月 3 13:50 0001.png
-rwxrwxr-x 1 jinzhongxu jinzhongxu 2169 3月 3 13:50 0002.png
-rwxrwxr-x 1 jinzhongxu jinzhongxu 2237 3月 3 13:50 0003.png
-rwxrwxr-x 1 jinzhongxu jinzhongxu 2104 3月 3 13:50 0004.png
data_dir = "/workspace/disk1/datasets/cifar10/cifar10/"
datasets_category = ["train", "test"]
train_dir = os.path.join(data_dir, datasets_category[0])
test_dir = os.path.join(data_dir, datasets_category[1])
# 数据输入模型前需要进行尺寸统一（都变为224x224），并进行一些数据增广和归一化
mean_vec = [0.485, 0.456, 0.406]
std_vec = [0.229, 0.224, 0.225]
data_transforms = {
datasets_category[0]: transforms.Compose(
[
transforms.Resize(256),
transforms.RandomRotation(45),
transforms.Resize(224),
transforms.RandomHorizontalFlip(p=0.5),
transforms.RandomVerticalFlip(p=0.5),
transforms.ColorJitter(
brightness=0.2, contrast=0.1, saturation=0.1, hue=0.1
),
transforms.RandomGrayscale(p=0.025),
transforms.ToTensor(),
transforms.Normalize(mean_vec, std_vec),
]
),
datasets_category[1]: transforms.Compose(
[
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean_vec, std_vec),
]
),
}
# 创建数据迭代器
batch_size = 8
image_datasets = {
x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x])
for x in datasets_category
}
dataloaders = {
x: DataLoader(image_datasets[x], batch_size=batch_size, shuffle=True)
for x in datasets_category
}
dataset_sizes = {x: len(image_datasets[x]) for x in datasets_category}
class_names = image_datasets[datasets_category[0]].classes
class_names
['airplane',
'automobile',
'bird',
'cat',
'deer',
'dog',
'frog',
'horse',
'ship',
'truck']
image_datasets
{'train': Dataset ImageFolder
Number of datapoints: 50000
Root location: /workspace/disk1/datasets/cifar10/cifar10/train
StandardTransform
Transform: Compose(
Resize(size=256, interpolation=bilinear, max_size=None, antialias=None)
RandomRotation(degrees=[-45.0, 45.0], interpolation=nearest, expand=False, fill=0)
Resize(size=224, interpolation=bilinear, max_size=None, antialias=None)
RandomHorizontalFlip(p=0.5)
RandomVerticalFlip(p=0.5)
ColorJitter(brightness=[0.8, 1.2], contrast=[0.9, 1.1], saturation=[0.9, 1.1], hue=[-0.1, 0.1])
RandomGrayscale(p=0.025)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
),
'test': Dataset ImageFolder
Number of datapoints: 10000
Root location: /workspace/disk1/datasets/cifar10/cifar10/test
StandardTransform
Transform: Compose(
Resize(size=256, interpolation=bilinear, max_size=None, antialias=None)
CenterCrop(size=(224, 224))
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)}
# 把 tensor 图像转为 numpy 图像，需要从数据预处理操作中逆映射回来
# 注意 tensor 的数据需要转换成 numpy 格式，而且还需要还原回标准化的结果
def im_convert(tensor):
image = tensor.to("cpu").clone().detach()
image = image.numpy().squeeze()
image = image.transpose(1, 2, 0)
image = image * np.array(mean_vec) + np.array(std_vec)
image = image.clip(0, 1)
return image
# 查看一下测试数据集
columns = 4
rows = 2
fig = plt.figure(figsize=(columns * 5, rows * 5))
dataiter = iter(dataloaders["test"])
inputs, classes = dataiter.next()
for idx in range(columns * rows):
ax = fig.add_subplot(rows, columns, idx + 1, xticks=[], yticks=[])
ax.set_title(class_names[classes[idx]])
plt.imshow(im_convert(inputs[idx]))
plt.show()
模型加载和训练
加载 models 中提供的模型，并且直接用训练好的权重当做初始化参数
因为模型参数多，而且都是已经训练好的参数，在我们的小数据集上（cifar10），我们先使用 resnet 模型的卷积层部分作为特征提取器，改变最后的全连接层从1000类别改为我们这里的10类
# 查看 resnet152 的模型结构
models.resnet152()
ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(6): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(7): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(6): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(7): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(8): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(9): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(10): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(11): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(12): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(13): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(14): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(15): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(16): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(17): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(18): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(19): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(20): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(21): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(22): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(23): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(24): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(25): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(26): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(27): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(28): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(29): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(30): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(31): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(32): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(33): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(34): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(35): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=1000, bias=True)
)
模型最后的全连接层的输出是1000类，需要更改为我们这里的10类
def set_parameter_requires_grad(model, feature_extract):
"""
如果只进行特征提取，则把模型的所有可学习参数‘冻结’，即参数不进行梯度传播
"""
if feature_extract:
for param in model.parameters():
param.requires_grad = False
def initialize_model(num_classes, feature_extract):
model = models.resnet152(weights=models.ResNet152_Weights.IMAGENET1K_V1)
# 是否冻结参数，如果固定，则只用来特征提取
set_parameter_requires_grad(model, feature_extract)
# 更换最后的全连接层，改为10类输出
num_fc_infeature = model.fc.in_features
model.fc = nn.Sequential(
nn.Linear(num_fc_infeature, num_classes), nn.LogSoftmax(dim=1)
)
return model
# 下载预训练好的模型，并进行模型结构修改
feature_extract = True
model = initialize_model(num_classes=10, feature_extract=feature_extract)
model = model.to(device)
# 模型保存
model_checkpoint = "/workspace/disk1/datasets/models/checkpoint_feature_extract.pth"
# 是否训练所有层
params_to_update = model.parameters()
print("需要训练的参数:")
if feature_extract:
params_to_update = []
for name, param in model.named_parameters():
if param.requires_grad == True:
params_to_update.append(param)
print("\t", name)
else:
for name, param in model.named_parameters():
if param.requires_grad == True:
print("\t", name)
需要训练的参数:
fc.0.weight
fc.0.bias
优化器设置
optimizer = optim.Adam(params_to_update, lr=1e-2)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1)
criterion = nn.NLLLoss()
def train_model(
model,
dataloaders,
criterion,
optimizer,
num_epochs=25,
model_checkpoint=None,
s="%Y年%m月%d日%H时%M分%S秒",
):
best_acc = 0
model.to(device)
train_acc_history = []
train_losses = []
test_acc_history = []
test_losses = []
LR = [optimizer.param_groups[0]["lr"]]
best_model_wts = copy.deepcopy(model.state_dict())
for epoch in range(num_epochs):
now = datetime.now().strftime(s)
print(f"------{now}------")
print(f"\033[1;34m Epoch {epoch+1}/{num_epochs}\033[0m")
# 先进行训练，然后再使用当前训练的模型进行测试
for phase in datasets_category:
if phase == "train":
model.train()
else:
model.eval()
running_loss = 0.0
running_corrects = 0
# 把所有数据都使用进行训练和测试
for inputs, labels in dataloaders[phase]:
inputs = inputs.to(device)
labels = labels.to(device)
# 记得先把梯度清零
optimizer.zero_grad()
# 只有训练的时候才计算和更新梯度
# 如果是测试阶段，则不设置需要计算梯度，也不进行反向创博
with torch.set_grad_enabled(phase == "train"): # 训练阶段也进行正向传播计算
outputs = model(inputs)
loss = criterion(outputs, labels)
_, preds = torch.max(outputs, 1) # 返回大于1的值和索引，这里只需要索引信息
# 训练阶段更新权重，测试阶段则不进行反向传播
if phase == "train":
loss.backward()
optimizer.step()
# 计算损失
running_loss += loss.item() * inputs.size(0)
running_corrects += torch.sum(preds == labels.data)
epoch_loss = running_loss / len(dataloaders[phase].dataset)
epoch_acc = running_corrects.double() / len(dataloaders[phase].dataset)
print(f"{phase.upper()}阶段: Loss: {epoch_loss:.4f}, Acc: {epoch_acc:.4f}")
# 保存每一个 epoch 得到的最好模型
if phase == "test" and epoch_acc > best_acc:
best_acc = epoch_acc
best_model_wts = copy.deepcopy(model.state_dict())
state = {
"state_dict": model.state_dict(),
"best_acc": best_acc,
"optimizer": optimizer.state_dict(),
}
torch.save(state, model_checkpoint)
if phase == "test":
test_acc_history.append(epoch_acc)
test_losses.append(epoch_loss)
scheduler.step()
if phase == "train":
train_acc_history.append(epoch_acc)
train_losses.append(epoch_loss)
print(f"当前学习率: {optimizer.param_groups[0]['lr']:.7f}")
LR.append(optimizer.param_groups[0]["lr"])
print()
# 训练完成
now = datetime.now().strftime(s)
print(f"------{now}------")
print(f"最好的测试准确率: {best_acc:.4f}")
# 返回最好的模型
model.load_state_dict(best_model_wts)
return model, test_acc_history, train_acc_history, test_losses, train_losses, LR
(
model,
test_acc_hisory,
train_acc_history,
test_losses,
train_losses,
LR,
) = train_model(
model=model,
dataloaders=dataloaders,
criterion=criterion,
optimizer=optimizer,
num_epochs=15,
model_checkpoint=model_checkpoint,
)
------2023年06月05日17时12分20秒------
[1;34m Epoch 1/15[0m
TRAIN阶段: Loss: 4.6413, Acc: 0.4249
TEST阶段: Loss: 4.9549, Acc: 0.5378
当前学习率: 0.0100000
------2023年06月05日17时24分14秒------
[1;34m Epoch 2/15[0m
TRAIN阶段: Loss: 4.9278, Acc: 0.4649
TEST阶段: Loss: 2.9911, Acc: 0.6334
当前学习率: 0.0100000
------2023年06月05日17时34分39秒------
[1;34m Epoch 3/15[0m
TRAIN阶段: Loss: 4.7944, Acc: 0.4789
TEST阶段: Loss: 3.4556, Acc: 0.6442
当前学习率: 0.0100000
------2023年06月05日17时45分12秒------
[1;34m Epoch 4/15[0m
TRAIN阶段: Loss: 4.7750, Acc: 0.4890
TEST阶段: Loss: 5.0836, Acc: 0.5772
当前学习率: 0.0100000
------2023年06月05日17时56分01秒------
[1;34m Epoch 5/15[0m
TRAIN阶段: Loss: 4.8759, Acc: 0.4900
TEST阶段: Loss: 4.2777, Acc: 0.6108
当前学习率: 0.0100000
------2023年06月05日18时08分10秒------
[1;34m Epoch 6/15[0m
TRAIN阶段: Loss: 4.8280, Acc: 0.4937
TEST阶段: Loss: 3.0909, Acc: 0.6849
当前学习率: 0.0100000
------2023年06月05日18时20分23秒------
[1;34m Epoch 7/15[0m
TRAIN阶段: Loss: 4.8318, Acc: 0.5012
TEST阶段: Loss: 4.9785, Acc: 0.6307
当前学习率: 0.0010000
------2023年06月05日18时31分24秒------
[1;34m Epoch 8/15[0m
TRAIN阶段: Loss: 2.2879, Acc: 0.6079
TEST阶段: Loss: 2.1652, Acc: 0.6859
当前学习率: 0.0010000
------2023年06月05日18时42分27秒------
[1;34m Epoch 9/15[0m
TRAIN阶段: Loss: 1.8844, Acc: 0.6099
TEST阶段: Loss: 2.0901, Acc: 0.6730
当前学习率: 0.0010000
------2023年06月05日18时53分30秒------
[1;34m Epoch 10/15[0m
TRAIN阶段: Loss: 1.7233, Acc: 0.6075
TEST阶段: Loss: 1.8010, Acc: 0.6860
当前学习率: 0.0010000
------2023年06月05日19时05分04秒------
[1;34m Epoch 11/15[0m
TRAIN阶段: Loss: 1.6624, Acc: 0.6058
TEST阶段: Loss: 1.8861, Acc: 0.6624
当前学习率: 0.0010000
------2023年06月05日19时17分24秒------
[1;34m Epoch 12/15[0m
TRAIN阶段: Loss: 1.5827, Acc: 0.6117
TEST阶段: Loss: 1.4209, Acc: 0.7079
当前学习率: 0.0010000
------2023年06月05日19时28分23秒------
[1;34m Epoch 13/15[0m
TRAIN阶段: Loss: 1.5545, Acc: 0.6092
TEST阶段: Loss: 1.5516, Acc: 0.6871
当前学习率: 0.0010000
------2023年06月05日19时39分20秒------
[1;34m Epoch 14/15[0m
TRAIN阶段: Loss: 1.5086, Acc: 0.6097
TEST阶段: Loss: 1.4248, Acc: 0.7011
当前学习率: 0.0001000
------2023年06月05日19时50分17秒------
[1;34m Epoch 15/15[0m
TRAIN阶段: Loss: 1.3143, Acc: 0.6375
TEST阶段: Loss: 1.5627, Acc: 0.6778
当前学习率: 0.0001000
------2023年06月05日20时01分21秒------
最好的测试准确率: 0.7079
再继续训练所有层
checkpoint = torch.load(model_checkpoint)
model.load_state_dict(checkpoint["state_dict"])
best_acc = checkpoint["best_acc"]
# optimizer.load_state_dict(checkpoint["optimizer"])
for param in model.parameters():
param.requires_grad = True
# 再继续训练所有的参数，学习率调小一点
optimizer = optim.Adam(model.parameters(), lr=1e-4)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1)
# 损失函数
criterion = nn.NLLLoss()
model_checkpoint = "/workspace/disk1/datasets/models/checkpoint_finetune.pth"
(
model,
test_acc_history,
train_acc_history,
test_losses,
train_losses,
LR,
) = train_model(
model=model,
dataloaders=dataloaders,
criterion=criterion,
optimizer=optimizer,
num_epochs=5,
model_checkpoint=model_checkpoint,
)
------2023年06月05日20时01分22秒------
[1;34m Epoch 1/5[0m
TRAIN阶段: Loss: 1.0279, Acc: 0.6982
TEST阶段: Loss: 0.5613, Acc: 0.8083
当前学习率: 0.0001000
------2023年06月05日20时22分24秒------
[1;34m Epoch 2/5[0m
TRAIN阶段: Loss: 0.6405, Acc: 0.7801
TEST阶段: Loss: 0.5291, Acc: 0.8192
当前学习率: 0.0001000
------2023年06月05日20时41分26秒------
[1;34m Epoch 3/5[0m
TRAIN阶段: Loss: 0.5285, Acc: 0.8168
TEST阶段: Loss: 0.4542, Acc: 0.8421
当前学习率: 0.0001000
------2023年06月05日21时00分25秒------
[1;34m Epoch 4/5[0m
TRAIN阶段: Loss: 0.4517, Acc: 0.8451
TEST阶段: Loss: 0.4239, Acc: 0.8558
当前学习率: 0.0001000
------2023年06月05日21时21分40秒------
[1;34m Epoch 5/5[0m
TRAIN阶段: Loss: 0.4008, Acc: 0.8616
TEST阶段: Loss: 0.3450, Acc: 0.8820
当前学习率: 0.0001000
------2023年06月05日21时40分49秒------
最好的测试准确率: 0.8820
加载训练好的模型
model = initialize_model(num_classes=10, feature_extract=feature_extract)
model = model.to(device)
checkpoint = torch.load(model_checkpoint)
best_acc = checkpoint["best_acc"]
model.load_state_dict(checkpoint["state_dict"])
<All keys matched successfully>
测试数据预处理
- 测试数据处理方法需要跟训练时一致
- crop 操作的目的是保证输入的大小一致
- 标注化操作是必须的，使用与训练数据相同的mean、std，训练数据是在0-1归一化的，测试数据也需要归一化
- pytorch 中颜色通道是第一个维度，需要转换
from PIL import Image
def preprocess_image(image_path):
img = Image.open(image_path)
if img.size[0] > img.size[1]:
img.thumbnail((10000, 256))
else:
img.thumbnail((256, 10000))
# crop
left_margin = (img.width - 224) / 2
bottom_margin = (img.height - 224) / 2
right_margin = left_margin + 224
top_margin = bottom_margin + 224
img = img.crop((left_margin, bottom_margin, right_margin, top_margin))
img = np.array(img) / 255
mean = np.array(mean_vec)
std = np.array(std_vec)
img = (img - mean) / std
img = img.transpose((2, 0, 1))
return img
def imshow(image, ax=None, title=None):
if ax is None:
fig, ax = plt.subplots()
image = np.array(image).transpose((1, 2, 0))
mean = np.array(mean_vec)
std = np.array(std_vec)
image = std * image + mean
image = np.clip(image, 0, 1)
ax.imshow(image)
ax.set_title(title)
return ax
image_path = "/workspace/disk1/datasets/cifar10/cifar10/train/airplane/0001.png"
img = preprocess_image(image_path)
imshow(img)
<AxesSubplot:>
img.shape
(3, 224, 224)
图像分类应用
# 得到一个 batch 的测试数据
dataiter = iter(dataloaders["test"])
images, labels = dataiter.next()
model.eval()
output = model(images.to(device))
# output 表示对一个 batch 中每一个数据得到其属于各个类别的可能性
output.shape
torch.Size([8, 10])
_, preds_tensor = torch.max(output, 1)
preds = (
np.squeeze(preds_tensor.numpy())
if not torch.cuda.is_available()
else np.squeeze(preds_tensor.cpu().numpy())
)
preds
array([7, 5, 0, 3, 2, 9, 5, 9])
columns = 4
rows = 2
fig = plt.figure(figsize=(columns * 5, rows * 5))
# 图片 title 中为 "预测值(ground truth)"，当预测错误时，使用红色标识
for idx in range(columns * rows):
ax = fig.add_subplot(rows, columns, idx + 1, xticks=[], yticks=[])
plt.imshow(im_convert(images[idx]))
ax.set_title(
f"{class_names[preds[idx]]} ({class_names[labels[idx].item()]})",
color="green" if preds[idx] == labels[idx].item() else "red",
)
plt.show()
Recommend
-
45
-
11
最近在学习一些检测方面的网络，使用的是pytorch。模型结构可视化是学习网络的有用的部分，pytorch没有原生支持这个功能，需要找一些其他方式，下面总结几种方法 （推荐用4） 。 1. torch .pt -> netron
-
60
修改PyTorch预训练模型的下载路径 PyTorch...
-
6
Pytorch入门教程16-Pytorch中模型的定义和参数初始化 2020年10月15日07:06:11
-
5
迁移学习之Domain Adaptation 发表于 ...
-
4
1.什么是JVM? 定义：Java Virtual Machine - java 程序的运行环境（java 二进制字节码的运行环境） 好处： 一次编写，到处运行 自动内存管理， 垃圾回收功能 数组下标越界检查
-
2
PGL图学习之图神经网络GNN模型GCN、GAT[系列六] 项目链接：一键fork直接跑程序 https://aistudi...
-
4
Train and Evaluate the Model 在创建了模型，使用您选择的优化器和损失函数对其进行编译，并对其内容进行了完整性检查之后，现在就可以开始构建了！ 只需调用.fit（）进行训练。就是这样！无需进行小批量...
-
7
其实这个笔记起源于一个报错，报错内容也很简单，希望传入一个三维的tensor，但是得到了一个四维。 RuntimeError: only batches of spatial targets supported (3...
-
3
使用Optuna进行PyTorch模型的超参数调优 作者：Ashish Verma 2023-06-06 15:42:13 开源 Optuna是一个开源的超参数优化框架，Optuna与...
About Joyk
Aggregate valuable and interesting links.
Joyk means Joy of geeK