昇思25天学习打卡营第8天 |昇思MindSpore ResNet50迁移学习

简介

在实际应用场景中，由于训练数据集不足，很少有人会从头开始训练整个网络。普遍做法是使用在大型数据集上预训练的模型，然后将其作为初始化权重或固定特征提取器，用于特定任务。本章使用迁移学习方法对ImageNet数据集中狼和狗图像进行分类。

数据准备

下载数据集
使用download接口下载狗与狼分类数据集，数据集目录结构如下：

datasets-Canidae/data/
└── Canidae
    ├── train
    │   ├── dogs
    │   └── wolves
    └── val
        ├── dogs
        └── wolves

加载数据集
使用mindspore.dataset.ImageFolderDataset接口来加载数据集，并进行图像增强操作。

import mindspore as ms
import mindspore.dataset as ds
import mindspore.dataset.vision as vision

batch_size = 18
image_size = 224
num_epochs = 5
lr = 0.001
momentum = 0.9
workers = 4

data_path_train = "./datasets-Canidae/data/Canidae/train/"
data_path_val = "./datasets-Canidae/data/Canidae/val/"

def create_dataset_canidae(dataset_path, usage):
    data_set = ds.ImageFolderDataset(dataset_path, num_parallel_workers=workers, shuffle=True)
    mean = [0.485 * 255, 0.456 * 255, 0.406 * 255]
    std = [0.229 * 255, 0.224 * 255, 0.225 * 255]
    scale = 32
    if usage == "train":
        trans = [
            vision.RandomCropDecodeResize(size=image_size, scale=(0.08, 1.0), ratio=(0.75, 1.333)),
            vision.RandomHorizontalFlip(prob=0.5),
            vision.Normalize(mean=mean, std=std),
            vision.HWC2CHW()
        ]
    else:
        trans = [
            vision.Decode(),
            vision.Resize(image_size + scale),
            vision.CenterCrop(image_size),
            vision.Normalize(mean=mean, std=std),
            vision.HWC2CHW()
        ]
    data_set = data_set.map(operations=trans, input_columns='image', num_parallel_workers=workers)
    data_set = data_set.batch(batch_size)
    return data_set

dataset_train = create_dataset_canidae(data_path_train, "train")
dataset_val = create_dataset_canidae(data_path_val, "val")

数据集可视化
通过create_dict_iterator接口创建数据迭代器，使用next迭代访问数据集，并进行可视化展示。

data = next(dataset_train.create_dict_iterator())
images = data["image"]
labels = data["label"]

import matplotlib.pyplot as plt
import numpy as np

class_name = {0: "dogs", 1: "wolves"}

plt.figure(figsize=(5, 5))
for i in range(4):
    data_image = images[i].asnumpy()
    data_label = labels[i]
    data_image = np.transpose(data_image, (1, 2, 0))
    mean = np.array([0.485, 0.456, 0.406])
    std = np.array([0.229, 0.224, 0.225])
    data_image = std * data_image + mean
    data_image = np.clip(data_image, 0, 1)
    plt.subplot(2, 2, i+1)
    plt.imshow(data_image)
    plt.title(class_name[int(labels[i].asnumpy())])
    plt.axis("off")
plt.show()

构建ResNet50模型

定义残差块
定义基础的ResidualBlockBase和扩展的ResidualBlock，包含卷积层、归一化层和ReLU激活函数。

搭建ResNet50网络
包括初始的卷积层、池化层、四个残差块组、平均池化层和全连接层。

from typing import Type, Union, List, Optional
from mindspore import nn, train
from mindspore.common.initializer import Normal

class ResidualBlockBase(nn.Cell):
    expansion = 1
    def __init__(self, in_channel, out_channel, stride=1, norm=None, down_sample=None):
        super(ResidualBlockBase, self).__init__()
        if not norm:
            self.norm = nn.BatchNorm2d(out_channel)
        else:
            self.norm = norm
        self.conv1 = nn.Conv2d(in_channel, out_channel, kernel_size=3, stride=stride, weight_init=Normal(mean=0, sigma=0.02))
        self.conv2 = nn.Conv2d(out_channel, out_channel, kernel_size=3, weight_init=Normal(mean=0, sigma=0.02))
        self.relu = nn.ReLU()
        self.down_sample = down_sample

    def construct(self, x):
        identity = x
        out = self.conv1(x)
        out = self.norm(out)
        out = self.relu(out)
        out = self.conv2(out)
        out = self.norm(out)
        if self.down_sample is not None:
            identity = self.down_sample(x)
        out += identity
        out = self.relu(out)
        return out

class ResidualBlock(nn.Cell):
    expansion = 4
    def __init__(self, in_channel, out_channel, stride=1, down_sample=None):
        super(ResidualBlock, self).__init__()
        self.conv1 = nn.Conv2d(in_channel, out_channel, kernel_size=1, weight_init=Normal(mean=0, sigma=0.02))
        self.norm1 = nn.BatchNorm2d(out_channel)
        self.conv2 = nn.Conv2d(out_channel, out_channel, kernel_size=3, stride=stride, weight_init=Normal(mean=0, sigma=0.02))
        self.norm2 = nn.BatchNorm2d(out_channel)
        self.conv3 = nn.Conv2d(out_channel, out_channel * self.expansion, kernel_size=1, weight_init=Normal(mean=0, sigma=0.02))
        self.norm3 = nn.BatchNorm2d(out_channel * self.expansion)
        self.relu = nn.ReLU()
        self.down_sample = down_sample

    def construct(self, x):
        identity = x
        out = self.conv1(x)
        out = self.norm1(out)
        out = self.relu(out)
        out = self.conv2(x)
        out = self.norm2(out)
        out = self.relu(out)
        out = self.conv3(out)
        out = self.norm3(out)
        if self.down_sample is not None:
            identity = self.down_sample(x)
        out += identity
        out = self.relu(out)
        return out

def make_layer(last_out_channel, block, channel, block_nums, stride=1):
    down_sample = None
    if stride != 1 or last_out_channel != channel * block.expansion:
        down_sample = nn.SequentialCell([
            nn.Conv2d(last_out_channel, channel * block.expansion, kernel_size=1, stride=stride, weight_init=Normal(mean=0, sigma=0.02)),
            nn.BatchNorm2d(channel * block.expansion, gamma_init=Normal(mean=1, sigma=0.02))
        ])
    layers = []
    layers.append(block(last_out_channel, channel, stride=stride, down_sample=down_sample))
    in_channel = channel * block.expansion
    for _ in range(1, block_nums):
        layers.append(block(in_channel, channel))
    return nn.SequentialCell(layers)

class ResNet(nn.Cell):
    def __init__(self, block, layer_nums, num_classes, input_channel):
        super(ResNet, self).__init__()
        self.relu = nn.ReLU()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, weight_init=Normal(mean=0, sigma=0.02))
        self.norm = nn.BatchNorm2d(64)
        self.max_pool = nn.MaxPool2d(kernel_size=3, stride=2, pad_mode='same')
        self.layer1 = make_layer(64, block, 64, layer_nums[0])
        self.layer2 = make_layer(64 * block.expansion, block, 128, layer_nums[1], stride=2)
        self.layer3 = make_layer(128 * block.expansion, block, 256, layer_nums[2], stride=2)
        self.layer4 = make_layer(256 * block.expansion, block, 512, layer_nums[3], stride=2)
        self.avg_pool = nn.AvgPool2d()
        self.flatten = nn.Flatten()
        self.fc = nn.Dense(in_channels=input_channel, out_channels=num_classes)

    def construct(self, x):
        x = self.conv1(x)
        x = self.norm(x)
        x = self.relu(x)
        x = self.max_pool(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.avg_pool(x)
        x = self.flatten(x)
        x = self.fc(x)
        return x

def _resnet(model_url, block, layers, num_classes

, input_channel):
    model = ResNet(block, layers, num_classes, input_channel)
    return model

def resnet50(num_classes=2, input_channel=2048):
    return _resnet("", ResidualBlock, [3, 4, 6, 3], num_classes, input_channel)

模型加载与冻结预训练权重
加载预训练的ResNet50模型权重，冻结部分网络参数以作为特征提取器。

from mindspore import load_checkpoint, load_param_into_net

def load_pretrained_weights(model, ckpt_path):
    param_dict = load_checkpoint(ckpt_path)
    load_param_into_net(model, param_dict)

model = resnet50()
pretrained_ckpt_path = "resnet50.ckpt"
load_pretrained_weights(model, pretrained_ckpt_path)

for param in model.get_parameters():
    param.requires_grad = False

微调网络
微调ResNet50的最后一个全连接层以适应新任务。

class ResNetFinetune(nn.Cell):
    def __init__(self, model, num_classes):
        super(ResNetFinetune, self).__init__()
        self.base_model = model
        self.fc = nn.Dense(2048, num_classes)

    def construct(self, x):
        x = self.base_model(x)
        x = self.fc(x)
        return x

finetune_model = ResNetFinetune(model, num_classes=2)

模型训练

定义训练参数和优化器，使用交叉熵损失函数。

import mindspore.nn as nn

loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
optimizer = nn.Momentum(finetune_model.trainable_params(), learning_rate=lr, momentum=momentum)

model = ms.Model(finetune_model, loss_fn, optimizer, metrics={"accuracy"})

训练和验证模型
进行模型训练并在验证集上进行评估。

model.train(num_epochs, dataset_train, dataset_sink_mode=False)
metric = model.eval(dataset_val, dataset_sink_mode=False)
print("Validation accuracy: ", metric['accuracy'])

总结

迁移学习方法能有效利用预训练模型在新任务上的表现，通过微调模型权重，可以在有限的数据上获得较好的性能。通过对ResNet50模型进行特定任务的微调，实现了对狼和狗图像的分类，提高了模型的泛化能力。