昇思25天学习打卡营第18天|ResNet50迁移学习

        在实际应用中，由于训练数据集不足，很少有人会从头开始训练整个网络。普遍做法是在一个大规模基础数据集上训练得到一个预训练模型，然后使用该模型来初始化网络的权重参数或作为固定特征提取器应用于特定任务中。本文将使用迁移学习的方法对ImageNet数据集中的狼和狗图像进行分类。

数据准备

加载数据集

        下载并解压狗与狼分类数据集，数据集中的图像来自ImageNet，每个分类有大约120张训练图像与30张验证图像。

%%capture captured_output
!pip uninstall mindspore -y
!pip install -i https://pypi.mirrors.ustc.edu.cn/simple mindspore==2.3.0rc1
!pip show mindspore
from download import download

dataset_url = "https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/notebook/datasets/intermediate/Canidae_data.zip"

download(dataset_url, "./datasets-Canidae", kind="zip", replace=True)

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

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")
step_size_train = dataset_train.get_dataset_size()

dataset_val = create_dataset_canidae(data_path_val, "val")
step_size_val = dataset_val.get_dataset_size()

构建模型

        本文使用ResNet50模型进行训练。通过将pretrained参数设置为True来下载ResNet50的预训练模型并将权重参数加载到网络中。

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

class ResidualBlockBase(nn.Cell):
    expansion = 1

    def __init__(self, in_channel, out_channel, stride=1, norm=None, down_sample=None):
        super(ResidualBlockBase, self).__init__()
        self.conv1 = nn.Conv2d(in_channel, out_channel, kernel_size=3, stride=stride, weight_init=Normal(0.02))
        self.norm1 = nn.BatchNorm2d(out_channel)
        self.relu = nn.ReLU()
        self.conv2 = nn.Conv2d(out_channel, out_channel, kernel_size=3, weight_init=Normal(0.02))
        self.norm2 = nn.BatchNorm2d(out_channel)
        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)
        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(0.02))
        self.norm1 = nn.BatchNorm2d(out_channel)
        self.conv2 = nn.Conv2d(out_channel, out_channel, kernel_size=3, stride=stride, weight_init=Normal(0.02))
        self.norm2 = nn.BatchNorm2d(out_channel)
        self.conv3 = nn.Conv2d(out_channel, out_channel * self.expansion, kernel_size=1, weight_init=Normal(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(out)
        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(0.02)),
            nn.BatchNorm2d(channel * block.expansion)
        ])
    layers = [block(last_out_channel, channel, stride, 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.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, weight_init=Normal(0.02))
        self.norm = nn.BatchNorm2d(64)
        self.relu = nn.ReLU()
        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, pretrained, pretrained_ckpt, input_channel):
    model = ResNet(block, layers, num_classes, input_channel)
    if pretrained:
        download(url=model_url, path=pretrained_ckpt, replace=True)
        param_dict = load_checkpoint(pretrained_ckpt)
        load_param_into_net(model, param_dict)
    return model

def resnet50(num_classes=1000, pretrained=False):
    resnet50_url = "https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/notebook/models/application/resnet50_224_new.ckpt"
    resnet50_ckpt = "./LoadPretrainedModel/resnet50_224_new.ckpt"
    return _resnet(resnet50_url, ResidualBlock, [3, 4, 6, 3], num_classes, pretrained, resnet50_ckpt, 2048)

固定特征进行训练

        使用固定特征进行训练的时候，需要冻结除最后一层之外的所有网络层。通过设置requires_grad=False冻结参数，以便不在反向传播中计算梯度。

import mindspore as ms
import matplotlib.pyplot as plt
import os
import time

net_work = resnet50(pretrained=True)

# 全连接层输入层的大小
in_channels = net_work.fc.in_channels
# 输出通道数大小为狼狗分类数2
head = nn.Dense(in_channels, 2)
# 重置全连接层
net_work.fc = head

# 平均池化层kernel size为7
avg_pool = nn.AvgPool2d(kernel_size=7)
# 重置平均池化层
net_work.avg_pool = avg_pool

# 冻结除最后一层外的所有参数
for param in net_work.get_parameters():
    if param.name not in ["fc.weight", "fc.bias"]:
        param.requires_grad = False

# 定义优化器和损失函数
opt = nn.Momentum(params=net_work.trainable_params(), learning_rate=lr, momentum=0.5)
loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')

def forward_fn(inputs, targets):
    logits = net_work(inputs)
    loss = loss_fn(logits, targets)
    return loss

grad_fn = ms.value_and_grad(forward_fn, None, opt.parameters)

def train_step(inputs, targets):
    loss, grads = grad_fn(inputs, targets)
    opt(grads)
    return loss

# 实例化模型
model1 = ms.train.Model(net_work, loss_fn, opt, metrics={"Accuracy": ms.train.Accuracy()})

训练和评估

        开始训练模型，与没有预训练模型相比，将节约一大半时间，因为此时可以不用计算部分梯度。保存评估精度最高的ckpt文件于当前路径的./BestCheckpoint/resnet50-best-freezing-param.ckpt。

import mindspore as ms
import matplotlib.pyplot as plt
import os
import time

dataset_train = create_dataset_canidae(data_path_train, "train")
step_size_train = dataset_train.get_dataset_size()

dataset_val = create_dataset_canidae(data_path_val, "val")
step_size_val = dataset_val.get_dataset_size()

num_epochs = 5

# 创建迭代器
data_loader_train = dataset_train.create_tuple_iterator(num_epochs=num_epochs)
data_loader_val = dataset_val.create_tuple_iterator(num_epochs=num_epochs)
best_ckpt_dir = "./BestCheckpoint"
best_ckpt_path = "./BestCheckpoint/resnet50-best-freezing-param.ckpt"

print("Start Training Loop ...")

best_acc = 0

for epoch in range(num_epochs):
    losses = []
    net_work.set_train()

    epoch_start = time.time()

    for i, (images, labels) in enumerate(data_loader_train):
        labels = labels.astype(ms.int32)
        loss = train_step(images, labels)
        losses.append(loss)

    acc = model1.eval(dataset_val)['Accuracy']

    epoch_end = time.time()
    epoch_seconds = (epoch_end - epoch_start) * 1000
    step_seconds = epoch_seconds/step_size_train

    print("-" * 20)
    print("Epoch: [%3d/%3d], Average Train Loss: [%5.3f], Accuracy: [%5.3f]" % (
        epoch+1, num_epochs, sum(losses)/len(losses), acc
    ))
    print("epoch time: %5.3f ms, per step time: %5.3f ms" % (
        epoch_seconds, step_seconds
    ))

    if acc > best_acc:
        best_acc = acc
        if not os.path.exists(best_ckpt_dir):
            os.mkdir(best_ckpt_dir)
        ms.save_checkpoint(net_work, best_ckpt_path)

print("=" * 80)
print(f"End of validation the best Accuracy is: {best_acc: 5.3f}, save the best ckpt file in {best_ckpt_path}", flush=True)

可视化模型预测

使用固定特征得到的best.ckpt文件对验证集的狼和狗图像数据进行预测。若预测字体为蓝色即为预测正确，若预测字体为红色则预测错误。

import matplotlib.pyplot as plt
import mindspore as ms

def visualize_model(best_ckpt_path, val_ds):
    net = resnet50()
    in_channels = net.fc.in_channels
    head = nn.Dense(in_channels, 2)
    net.fc = head
    avg_pool = nn.AvgPool2d(kernel_size=7)
    net.avg_pool = avg_pool
    param_dict = ms.load_checkpoint(best_ckpt_path)
    ms.load_param_into_net(net, param_dict)
    model = ms.train.Model(net)
    data = next(val_ds.create_dict_iterator())
    images = data["image"].asnumpy()
    labels = data["label"].asnumpy()
    class_name = {0: "dogs", 1: "wolves"}
    output = model.predict(ms.Tensor(data['image']))
    pred = np.argmax(output.asnumpy(), axis=1)

    plt.figure(figsize=(5, 5))
    for i in range(4):
        plt.subplot(2, 2, i + 1)
        color = 'blue' if pred[i] == labels[i] else 'red'
        plt.title('predict:{}'.format(class_name[pred[i]]), color=color)
        picture_show = np.transpose(images[i], (1, 2, 0))
        mean = np.array([0.485, 0.456, 0.406])
        std = np.array([0.229, 0.224, 0.225])
        picture_show = std * picture_show + mean
        picture_show = np.clip(picture_show, 0, 1)
        plt.imshow(picture_show)
        plt.axis('off')

    plt.show()

visualize_model(best_ckpt_path, dataset_val)

结果

学习心得：通过这次使用迁移学习进行图像分类的学习，我进一步理解了如何利用预训练模型来提升模型性能并减少训练时间。在数据量有限的情况下，迁移学习能有效地利用在大规模数据集上训练好的模型，迅速提高特定任务的性能。在实现ResNet50的过程中，通过构建残差模块和堆叠残差块，更深刻地理解了ResNet网络的架构设计及其解决退化问题的优势。此外，在使用预训练模型进行微调时，通过冻结部分网络层的参数，可以有效避免过拟合，并节省计算资源。

如果你觉得这篇博文对你有帮助，请点赞、收藏、关注我，并且可以打赏支持我！

欢迎关注我的后续博文，我将分享更多关于深度学习和计算机视觉的精彩内容。

谢谢大家的支持！