基于MindSpore的恶性皮肤肿瘤识别
第一节：系统架构及神经网络讲解
第二节：系统实现流程
1 实验介绍
本实验旨在使用MindSpore框架在包含4个类别的数据集上，进行模型微调以训练ResNet50模型，实现皮肤病识别模型；
我们将利用MindSpore model_zoo中提供的ResNet50模型定义，通过昇思大模型平台进行模型的训练和优化；
1.1 实验目标
- 掌握使用MindSpore框架进行深度学习模型训练的基本流程；
- 理解ResNet50模型结构及其在图像分类任务中的应用；
- 学习如何针对特定数据集（皮肤病图像）调整和优化模型参数；
1.2 使用工具介绍
- MindSpore框架：开源深度学习框架，支持端、边、云多种场景，旨在为用户提供全场景AI解决方案；
- ResNet50模型：深度残差网络，通过引入残差学习解决了深度网络训练中的梯度消失问题，是图像识别领域广泛使用的模型之一；
- 昇思大模型平台：一个基于昇思Mindspore的深度学习平台，提供强大的计算资源和丰富的模型库；
1.3 实验步骤
- 数据准备：收集并预处理皮肤病图像数据集，确保数据质量满足训练需求。
- 模型选择：使用MindSpore model_zoo中的ResNet50模型作为基础模型。
- 模型训练：在昇思大模型平台上配置训练参数，启动模型训练过程。
- 性能评估：通过验证集评估模型性能，调整超参数以优化模型。
- 模型部署：将训练好的模型部署到实际应用中，进行皮肤病识别。
1.4 预备知识
- 熟练使用Python，了解Shell及Linux操作系统基本知识；
- 具备一定的深度学习理论知识，如卷积神经网络、损失函数、优化器，训练策略、Checkpoint等；
- 了解并熟悉MindSpore AI框架，MindSpore官网：昇思MindSpore | 全场景AI框架 | 昇思MindSpore社区官网
%%capture captured_output
# 实验环境已经预装了mindspore==2.3.0，如需更换mindspore版本，可更改下面 MINDSPORE_VERSION 变量
!pip uninstall mindspore -y
%env MINDSPORE_VERSION=2.3.0
!pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/${MINDSPORE_VERSION}/MindSpore/unified/aarch64/mindspore-${MINDSPORE_VERSION}-cp39-cp39-linux_aarch64.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.mirrors.ustc.edu.cn/simple# 查看当前 mindspore 版本
!pip show mindsporeName: mindspore Version: 2.3.0 Summary: MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios. Home-page: https://www.mindspore.cn Author: The MindSpore Authors Author-email: contact@mindspore.cn License: Apache 2.0 Location: /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages Requires: asttokens, astunparse, numpy, packaging, pillow, protobuf, psutil, scipy Required-by:
2 数据集处理
2.1 数据集介绍
数据集共有四类皮肤图片，分别为Basal Cell Carcinoma（基底细胞癌）、Melanoma（黑色素瘤）、Pigmented Benign Keratosis（色素性良性角化病）、Nevus（痣）四种；
- 基底细胞癌：是一种常见的皮肤癌，起源于表皮的基底细胞层，通常不会转移到身体其他部位，但如果不治疗，可能会局部侵袭并破坏周围组织；
- 黑色素瘤：是一种较为严重的皮肤癌，起源于皮肤中的黑色素细胞，可以迅速扩散到身体其他部位，因此早期发现和治疗至关重要；
- 痣：为色素痣，是皮肤上的一种色素沉着区域，通常为良性，可以是先天性的，也可以是后天发展的；
- 色素性良性角化病：是一种良性皮肤病变，表现为皮肤上的小色素斑块，通常不会引起健康问题，但有时可能会被误认为是黑色素瘤；
2.2 数据集获取和整理
数据集这里已经上传到昇思大模型社区内，链接为：昇思大模型平台
这里可以用git快速获取，下载后存储在pifudata_Maker文件夹内
!git clone https://source-xihe-mindspore.osinfra.cn/knoka/pifudata_Maker.gitCloning into 'pifudata_Maker'... remote: Enumerating objects: 12, done. remote: Counting objects: 100% (12/12), done. remote: Compressing objects: 100% (9/9), done. remote: Total 12 (delta 0), reused 0 (delta 0), pack-reused 0 Receiving objects: 100% (12/12), 462.96 MiB | 14.32 MiB/s, done.
%%capture captured_output
# 解压数据集pifudata_Maker/skin-cancer-detection.tar 并存储到pifudata_Maker文件夹内
!tar -xv -f pifudata_Maker/skin-cancer-detection.tar -C pifudata_Maker/下载后可以看到我们的数据集为如下形式，其中dataset_dir 根目录名称可以根据需要进行更改。
训练图像目录 train 包含用于训练的各个皮肤图像类别的目录
class1class2class3class4
验证图像目录 val 包含用于验证的各个皮瘤图像类别的目录
class1class2class3class4
下面部分代码则是将文件夹英文名转为数字0-3命名，至于为什么为这种格式，则源由于图像分类任务通常为将数据集组织成文件夹，其中每个文件夹的名称代表一个类别；
因此会期望每个文件夹名称是一个可以识别的标签，如果文件夹名称是数字，则可以将其直接作为类别标签，因为在训练过程中模型需要将输入数据（图像）与输出标签（类别）关联起来；
将疾病名称重命名为对应的编号后，MindSpore提供一种简便的方式来加载和处理数据集，即通过文件夹名称自动将图片分配到对应的类别；因此确保数据集的转换如此，可以简化数据加载和处理过程，直接使用文件夹名称（即编号）作为图像的标签，而不需要额外的映射步骤。
import os# 定义疾病名称和编号的映射
diseases = {0: "basal_cell_carcinoma",1: "melanoma",2: "nevus",3: "pigmented_benign_keratosis",
}# 定义目标文件夹路径
target_directory = "pifudata_Maker/images/" # 替换为你的文件夹路径# 遍历每个疾病名称和编号
for folder in ["train", "val"]:folder_path = os.path.join(target_directory, folder)print(folder_path)for number,disease in diseases.items():# 构造完整的旧文件夹路径old_folder_path = os.path.join(folder_path, disease)# 检查旧文件夹是否存在if not os.path.exists(old_folder_path):print(f"Folder not found: {old_folder_path}")continue# 构造新的文件夹名称new_folder_name = f"{number}"# 构造完整的新文件夹路径new_folder_path = os.path.join(folder_path, new_folder_name)# 重命名文件夹os.rename(old_folder_path, new_folder_path)print(f"Renamed {old_folder_path} to {new_folder_path}")pifudata_Maker/images/train Renamed pifudata_Maker/images/train/basal_cell_carcinoma to pifudata_Maker/images/train/0 Renamed pifudata_Maker/images/train/melanoma to pifudata_Maker/images/train/1 Renamed pifudata_Maker/images/train/nevus to pifudata_Maker/images/train/2 Renamed pifudata_Maker/images/train/pigmented_benign_keratosis to pifudata_Maker/images/train/3 pifudata_Maker/images/val Renamed pifudata_Maker/images/val/basal_cell_carcinoma to pifudata_Maker/images/val/0 Renamed pifudata_Maker/images/val/melanoma to pifudata_Maker/images/val/1 Renamed pifudata_Maker/images/val/nevus to pifudata_Maker/images/val/2 Renamed pifudata_Maker/images/val/pigmented_benign_keratosis to pifudata_Maker/images/val/3
2.3 数据集比例可视化
通过对数据集进行可视化展示，便于对数据进行直观的查看，为后续超参数的设置等做出帮助；
共有Basal Cell Carcinoma（基底细胞癌）、Melanoma（黑色素瘤）、Pigmented Benign Keratosis（色素性良性角化病）、Nevus（痣）四种皮肤；
其中Nevus训练集图片为357张、Melanoma训练集图片为438张、Pigmented Benign Keratosis训练集图片为462张、Basal Cell Carcinoma训练集图片为376张，四者验证集均为16张；
import os
import matplotlib.pyplot as plt# 疾病编号到名称的映射
diseases = {0: "basal_cell_carcinoma",1: "melanoma",2: "nevus",3: "pigmented_benign_keratosis",
}# 数据集目录路径
data_path_train = "pifudata_Maker/images/train"
data_path_val = "pifudata_Maker/images/val"# 初始化一个字典来存储每个类别的文件数量
file_counts_train = {}
file_counts_val = {}# 遍历训练数据文件夹
for folder_name in os.listdir(data_path_train):folder_path = os.path.join(data_path_train, folder_name)# 确保是文件夹if os.path.isdir(folder_path):# 计算文件夹内的文件数量count = len(os.listdir(folder_path))# 使用疾病编号作为字典的键file_counts_train[diseases.get(folder_name, folder_name)] = count# 遍历验证数据文件夹
for folder_name in os.listdir(data_path_val):folder_path = os.path.join(data_path_val, folder_name)# 确保是文件夹if os.path.isdir(folder_path):# 计算文件夹内的文件数量count = len(os.listdir(folder_path))# 使用疾病编号作为字典的键file_counts_val[diseases.get(folder_name, folder_name)] = count# 获取训练集的疾病名称列表
train_categories = list(file_counts_train.keys())# 使用matplotlib生成条形图
plt.figure(figsize=(12, 8))# 绘制训练集的条形图
train_bars = plt.bar(train_categories, file_counts_train.values(), width=0.4, color='blue', label='Train')# 绘制验证集的条形图，错开位置
val_bars = plt.bar([train_categories.index(cat) + 0.4 for cat in train_categories], [file_counts_val.get(cat, 0) for cat in train_categories], width=0.4, color='green', label='Validation')# 添加图例
plt.legend()# 使用疾病名称作为X轴标签
plt.xlabel('labels')
plt.ylabel('Number of Images')
plt.title('Training and Validation Datasets')print(train_categories)train_list = [diseases[int(i)] for i in train_categories]
print(train_list)plt.xticks([train_categories.index(cat) + 0.2 for cat in train_categories], [diseases[int(i)] for i in train_categories], rotation=30,fontsize='small')# 在柱形条上面显示数量
for i, (train_bar, val_bar) in enumerate(zip(train_bars, val_bars)):train_height = train_bar.get_height()val_height = val_bar.get_height() if val_bar else 0plt.text(train_bar.get_x() + train_bar.get_width() / 2, train_height, '{}'.format(train_height), ha='center', va='bottom')plt.text(val_bar.get_x() + val_bar.get_width() / 2, val_height, '{}'.format(val_height), ha='center', va='bottom') if val_bar else Noneplt.show()['2', '1', '0', '3'] ['nevus', 'melanoma', 'basal_cell_carcinoma', 'pigmented_benign_keratosis']
2.4 数据加载
-
使用MindSpore框架的mindspore.dataset模块来加载和预处理图像数据集，以便进行模型训练和验证；
-
定义数据集的路径，并创建了一个函数来加载和应用数据增强操作（如随机裁剪、随机水平翻转等），应用于训练集以提高模型泛化能力，验证集进行简单的解码、调整大小和归一化操作；
-
将图像数据集分批处理，以便于训练过程中的批处理，打印出训练集和验证集的批次数（即数据集大小除以批量大小），并展示验证集中第一个批次图像的张量形状和标签；
# 超参数定义
batch_size = 64 # 批量大小
image_size = 224 # 训练图像空间大小
num_epochs = 25 # 训练周期数
lr = 0.001 # 学习率
momentum = 0.6 # 动量
workers = 4 # 并行线程个数
import mindspore
mindspore.set_context(device_target="Ascend")import mindspore as ms
import mindspore.dataset as ds
import mindspore.dataset.vision as vision# 数据集目录路径
data_path_train = "pifudata_Maker/images/train"
data_path_val = "pifudata_Maker/images/val"# 创建训练数据集def create_dataset_canidae(dataset_path, usage):"""数据加载"""data_set = ds.ImageFolderDataset(dataset_path,num_parallel_workers=workers,shuffle=True,)# 数据增强操作mean = [0.485 * 255, 0.456 * 255, 0.406 * 255]std = [0.229 * 255, 0.224 * 255, 0.225 * 255]scale = 32if usage == "train":# Define map operations for training datasettrans = [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:# Define map operations for inference datasettrans = [vision.Decode(),vision.Resize(image_size + scale),vision.CenterCrop(image_size),vision.Normalize(mean=mean, std=std),vision.HWC2CHW()]# 数据映射操作data_set = data_set.map(operations=trans,input_columns='image',num_parallel_workers=workers)# 批量操作data_set = data_set.batch(batch_size)return data_setdataset_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()
print(step_size_train)
print(step_size_val)
data = next(dataset_val.create_dict_iterator())
images = data["image"]
labels = data["label"]
# print(data["image"][0])
print("Tensor of image", images.shape)
print("Labels:", labels)26 1 Tensor of image (64, 3, 224, 224) Labels: [2 2 1 1 1 2 3 0 0 3 0 1 0 3 0 0 3 0 2 2 3 0 3 1 2 3 0 2 1 1 0 3 1 2 3 0 12 1 0 3 2 0 1 3 3 1 0 2 2 3 2 0 0 2 3 1 1 2 1 2 3 1 3]
2.5 数据集可视化
为了更便于理解和调试代码，我们使用matplotlib库和numpy库来可视化数据集中的前四个图像。
import matplotlib.pyplot as plt
import numpy as np# class_name对应label，按文件夹字符串从小到大的顺序标记label
class_name = {0: "basal_cell_carcinoma",1: "melanoma",2: "nevus",3: "pigmented_benign_keratosis",
}# print(images[0])
plt.figure(figsize=(5, 5))
for i in range(4):# 获取图像及其对应的label# print(images[i])data_image = images[i].asnumpy()data_label = labels[i]# 处理图像供展示使用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 + meandata_image = np.clip(data_image, 0, 1)# 显示图像plt.subplot(2, 2, i+1)plt.imshow(data_image)plt.title(class_name[int(labels[i].asnumpy())])plt.axis("off")plt.show()
3 模型训练
3.1 网络搭建
在MindSpore中构建和配置深度残差网络，用于图像识别等任务，通过代码定义一个基于MindSpore框架的ResNet（残差网络）模型，包括基本的残差块（ResidualBlockBase和ResidualBlock）和完整的ResNet类；
提供_resnet和resnet50函数来实例化ResNet模型，加载预训练权重；
from typing import Type, Union, List, Optional
from mindspore import nn, train
from mindspore.common.initializer import Normalweight_init = Normal(mean=0, sigma=0.02)
gamma_init = Normal(mean=1, sigma=0.02)
class ResidualBlockBase(nn.Cell):expansion: int = 1 # 最后一个卷积核数量与第一个卷积核数量相等def __init__(self, in_channel: int, out_channel: int,stride: int = 1, norm: Optional[nn.Cell] = None,down_sample: Optional[nn.Cell] = None) -> None:super(ResidualBlockBase, self).__init__()if not norm:self.norm = nn.BatchNorm2d(out_channel)else:self.norm = normself.conv1 = nn.Conv2d(in_channel, out_channel,kernel_size=3, stride=stride,weight_init=weight_init)self.conv2 = nn.Conv2d(in_channel, out_channel,kernel_size=3, weight_init=weight_init)self.relu = nn.ReLU()self.down_sample = down_sampledef construct(self, x):"""ResidualBlockBase construct."""identity = x # shortcuts分支out = self.conv1(x) # 主分支第一层：3*3卷积层out = self.norm(out)out = self.relu(out)out = self.conv2(out) # 主分支第二层：3*3卷积层out = self.norm(out)if self.down_sample is not None:identity = self.down_sample(x)out += identity # 输出为主分支与shortcuts之和out = self.relu(out)return out
class ResidualBlock(nn.Cell):expansion = 4 # 最后一个卷积核的数量是第一个卷积核数量的4倍def __init__(self, in_channel: int, out_channel: int,stride: int = 1, down_sample: Optional[nn.Cell] = None) -> None:super(ResidualBlock, self).__init__()self.conv1 = nn.Conv2d(in_channel, out_channel,kernel_size=1, weight_init=weight_init)self.norm1 = nn.BatchNorm2d(out_channel)self.conv2 = nn.Conv2d(out_channel, out_channel,kernel_size=3, stride=stride,weight_init=weight_init)self.norm2 = nn.BatchNorm2d(out_channel)self.conv3 = nn.Conv2d(out_channel, out_channel * self.expansion,kernel_size=1, weight_init=weight_init)self.norm3 = nn.BatchNorm2d(out_channel * self.expansion)self.relu = nn.ReLU()self.down_sample = down_sampledef construct(self, x):identity = x # shortscuts分支out = self.conv1(x) # 主分支第一层：1*1卷积层out = self.norm1(out)out = self.relu(out)out = self.conv2(out) # 主分支第二层：3*3卷积层out = self.norm2(out)out = self.relu(out)out = self.conv3(out) # 主分支第三层：1*1卷积层out = self.norm3(out)if self.down_sample is not None:identity = self.down_sample(x)out += identity # 输出为主分支与shortcuts之和out = self.relu(out)return out
def make_layer(last_out_channel, block: Type[Union[ResidualBlockBase, ResidualBlock]],channel: int, block_nums: int, stride: int = 1):down_sample = None # shortcuts分支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=weight_init),nn.BatchNorm2d(channel * block.expansion, gamma_init=gamma_init)])layers = []layers.append(block(last_out_channel, channel, stride=stride, down_sample=down_sample))in_channel = channel * block.expansion# 堆叠残差网络for _ in range(1, block_nums):layers.append(block(in_channel, channel))return nn.SequentialCell(layers)
from mindspore import load_checkpoint, load_param_into_netclass ResNet(nn.Cell):def __init__(self, block: Type[Union[ResidualBlockBase, ResidualBlock]],layer_nums: List[int], num_classes: int, input_channel: int) -> None:super(ResNet, self).__init__()self.relu = nn.ReLU()# 第一个卷积层，输入channel为3（彩色图像），输出channel为64self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, weight_init=weight_init)self.norm = nn.BatchNorm2d(64)# 最大池化层，缩小图片的尺寸self.max_pool = nn.MaxPool2d(kernel_size=3, stride=2, pad_mode='same')# 各个残差网络结构块定义，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)# 平均池化层self.avg_pool = nn.AvgPool2d()# flattern层self.flatten = nn.Flatten()# 全连接层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 xdef _resnet(model_url: str, block: Type[Union[ResidualBlockBase, ResidualBlock]],layers: List[int], num_classes: int, pretrained: bool, pretrianed_ckpt: str,input_channel: int):model = ResNet(block, layers, num_classes, input_channel)if pretrained:# 加载预训练模型download(url=model_url, path=pretrianed_ckpt, replace=True)param_dict = load_checkpoint(pretrianed_ckpt)load_param_into_net(model, param_dict)return modeldef resnet50(num_classes: int = 1000, pretrained: bool = False):"ResNet50模型"resnet50_url = "https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/notebook/models/application/resnet50_224_new.ckpt"resnet50_ckpt = "./pretrainmodel/resnet50_224_new.ckpt"return _resnet(resnet50_url, ResidualBlock, [3, 4, 6, 3], num_classes,pretrained, resnet50_ckpt, 2048)3.2 模型训练
使用MindSpore框架在NPU上训练一个自定义的ResNet50模型，该模型被修改为用于一个4类的分类任务；
定义优化器、损失函数和评估指标，实例化训练模型，选择最佳准确率以及最佳模型检查点路径保存；
--------------------
Epoch: [ 23/ 25], Average Train Loss: [0.418], Accuracy: [0.734]
epoch time: 4133.578 ms, per step time: 158.984 ms
--------------------
Epoch: [ 24/ 25], Average Train Loss: [0.448], Accuracy: [0.719]
epoch time: 3491.099 ms, per step time: 134.273 ms
--------------------
Epoch: [ 25/ 25], Average Train Loss: [0.390], Accuracy: [0.719]
epoch time: 3515.689 ms, per step time: 135.219 ms
================================================================================
End of validation the best Accuracy is: 0.734, save the best ckpt file in ./BestCheckpoint/resnet50-best.ckpt# 代码中首先导入了必要的MindSpore模块和函数，并设置了运行环境。
import mindspore
from mindspore import nn, train
from mindspore.nn import Loss, Accuracy
!pip install download
import mindspore as ms
from download import downloadnetwork = resnet50(pretrained=True)# 通过替换ResNet50的原始全连接层和平均池化层来适配新的任务
# 全连接层输入层的大小
in_channels = network.fc.in_channels
# 输出通道数大小为皮肤肿瘤分类数4
head = nn.Dense(in_channels, 4)
# 重置全连接层
network.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
network.avg_pool = avg_poolimport mindspore as ms
import mindspore# 定义优化器和损失函数
opt = nn.Momentum(params=network.trainable_params(), learning_rate=lr, momentum=momentum)
loss_fn = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')# 实例化模型
model = train.Model(network, loss_fn, opt, metrics={"Accuracy": Accuracy()})def forward_fn(inputs, targets):logits = network(inputs)loss = loss_fn(logits, targets)return lossgrad_fn = mindspore.ops.value_and_grad(forward_fn, None, opt.parameters)def train_step(inputs, targets):loss, grads = grad_fn(inputs, targets)opt(grads)return loss# 创建迭代器
data_loader_train = dataset_train.create_tuple_iterator(num_epochs=num_epochs)
# 最佳模型保存路径
best_ckpt_dir = "./BestCheckpoint"
best_ckpt_path = "./BestCheckpoint/resnet50-best.ckpt"
import os
import time# 开始循环训练
print("Start Training Loop ...")best_acc = 0# 训练循环中，数据通过迭代器被加载，模型在每个epoch中更新权重，并计算训练损失。
# 在每个epoch结束时，模型在验证集上评估准确率，并保存具有最高准确率的模型检查点。
for epoch in range(num_epochs):losses = []network.set_train()epoch_start = time.time()# 为每轮训练读入数据for i, (images, labels) in enumerate(data_loader_train):labels = labels.astype(ms.int32)loss = train_step(images, labels)losses.append(loss)# 每个epoch结束后，验证准确率acc = model.eval(dataset_val)['Accuracy']epoch_end = time.time()epoch_seconds = (epoch_end - epoch_start) * 1000step_seconds = epoch_seconds/step_size_trainprint("-" * 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 = accif not os.path.exists(best_ckpt_dir):os.mkdir(best_ckpt_dir)ms.save_checkpoint(network, best_ckpt_path)print("=" * 80)
print(f"End of validation the best Accuracy is: {best_acc: 5.3f}, "f"save the best ckpt file in {best_ckpt_path}", flush=True)Looking in indexes: Simple Index Requirement already satisfied: download in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (0.3.5) Requirement already satisfied: tqdm in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from download) (4.66.5) Requirement already satisfied: six in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from download) (1.16.0) Requirement already satisfied: requests in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from download) (2.32.3) Requirement already satisfied: charset-normalizer<4,>=2 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (3.3.2) Requirement already satisfied: idna<4,>=2.5 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (3.10) Requirement already satisfied: urllib3<3,>=1.21.1 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (2.2.3) Requirement already satisfied: certifi>=2017.4.17 in /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages (from requests->download) (2024.8.30) Downloading data from https://mindspore-website.obs.cn-north-4.myhuaweicloud.com/notebook/models/application/resnet50_224_new.ckpt (97.7 MB)file_sizes: 100%|█████████████████████████████| 102M/102M [00:00<00:00, 113MB/s] Successfully downloaded file to ./pretrainmodel/resnet50_224_new.ckpt Start Training Loop ... -/usr/local/Ascend/ascend-toolkit/latest/opp/built-in/op_impl/ai_core/tbe/impl/util/util_conv2d_dynamic.py:130: UserWarning: conv2d fmap ori_range changed from [[32, 2147483647], [2048, 2048], [7, 15], [7, 15]] to [[32, 2147483647], [2048, 2048], [7, 15], (7, 15)].warnings.warn(to_print)-------------------- Epoch: [ 1/ 25], Average Train Loss: [1.353], Accuracy: [0.344] epoch time: 101835.918 ms, per step time: 3916.766 ms -------------------- Epoch: [ 2/ 25], Average Train Loss: [1.251], Accuracy: [0.484] epoch time: 3650.835 ms, per step time: 140.417 ms -------------------- Epoch: [ 3/ 25], Average Train Loss: [1.142], Accuracy: [0.453] epoch time: 3632.370 ms, per step time: 139.707 ms -------------------- Epoch: [ 4/ 25], Average Train Loss: [1.029], Accuracy: [0.516] epoch time: 3619.441 ms, per step time: 139.209 ms -------------------- Epoch: [ 5/ 25], Average Train Loss: [0.945], Accuracy: [0.531] epoch time: 3628.954 ms, per step time: 139.575 ms -------------------- Epoch: [ 6/ 25], Average Train Loss: [0.867], Accuracy: [0.562] epoch time: 3697.904 ms, per step time: 142.227 ms -------------------- Epoch: [ 7/ 25], Average Train Loss: [0.808], Accuracy: [0.609] epoch time: 3713.705 ms, per step time: 142.835 ms -------------------- Epoch: [ 8/ 25], Average Train Loss: [0.756], Accuracy: [0.625] epoch time: 3771.178 ms, per step time: 145.045 ms -------------------- Epoch: [ 9/ 25], Average Train Loss: [0.718], Accuracy: [0.641] epoch time: 3764.896 ms, per step time: 144.804 ms -------------------- Epoch: [ 10/ 25], Average Train Loss: [0.677], Accuracy: [0.609] epoch time: 3730.324 ms, per step time: 143.474 ms -------------------- Epoch: [ 11/ 25], Average Train Loss: [0.631], Accuracy: [0.641] epoch time: 3760.073 ms, per step time: 144.618 ms -------------------- Epoch: [ 12/ 25], Average Train Loss: [0.625], Accuracy: [0.688] epoch time: 3675.349 ms, per step time: 141.360 ms -------------------- Epoch: [ 13/ 25], Average Train Loss: [0.579], Accuracy: [0.672] epoch time: 3675.514 ms, per step time: 141.366 ms -------------------- Epoch: [ 14/ 25], Average Train Loss: [0.580], Accuracy: [0.703] epoch time: 3654.391 ms, per step time: 140.553 ms -------------------- Epoch: [ 15/ 25], Average Train Loss: [0.540], Accuracy: [0.656] epoch time: 3706.080 ms, per step time: 142.542 ms -------------------- Epoch: [ 16/ 25], Average Train Loss: [0.540], Accuracy: [0.719] epoch time: 3715.602 ms, per step time: 142.908 ms -------------------- Epoch: [ 17/ 25], Average Train Loss: [0.531], Accuracy: [0.719] epoch time: 3737.168 ms, per step time: 143.737 ms -------------------- Epoch: [ 18/ 25], Average Train Loss: [0.488], Accuracy: [0.719] epoch time: 3771.855 ms, per step time: 145.071 ms -------------------- Epoch: [ 19/ 25], Average Train Loss: [0.490], Accuracy: [0.703] epoch time: 3730.419 ms, per step time: 143.478 ms -------------------- Epoch: [ 20/ 25], Average Train Loss: [0.479], Accuracy: [0.750] epoch time: 3715.534 ms, per step time: 142.905 ms -------------------- Epoch: [ 21/ 25], Average Train Loss: [0.455], Accuracy: [0.719] epoch time: 3695.081 ms, per step time: 142.119 ms -------------------- Epoch: [ 22/ 25], Average Train Loss: [0.451], Accuracy: [0.750] epoch time: 3741.154 ms, per step time: 143.891 ms -------------------- Epoch: [ 23/ 25], Average Train Loss: [0.448], Accuracy: [0.766] epoch time: 3727.470 ms, per step time: 143.364 ms -------------------- Epoch: [ 24/ 25], Average Train Loss: [0.408], Accuracy: [0.719] epoch time: 3673.554 ms, per step time: 141.291 ms -------------------- Epoch: [ 25/ 25], Average Train Loss: [0.416], Accuracy: [0.766] epoch time: 3663.796 ms, per step time: 140.915 ms ================================================================================ End of validation the best Accuracy is: 0.766, save the best ckpt file in ./BestCheckpoint/resnet50-best.ckpt
3.3 模型评估
从验证数据集中提取了一批次的数据，使用模型进行预测，并根据预测结果将图像及其预测标签以蓝色（正确）或红色（错误）显示。
使用matplotlib库绘制了一个包含四个子图的图像，每个子图展示了一张图像及其预测结果。
import matplotlib.pyplot as plt
import mindspore as msdef visualize_model(best_ckpt_path, val_ds):net = resnet50()# 全连接层输入层的大小in_channels = net.fc.in_channels# 输出通道数大小为分类数4head = nn.Dense(in_channels, 4)# 重置全连接层net.fc = head# 平均池化层kernel size为7avg_pool = nn.AvgPool2d(kernel_size=7)# 重置平均池化层net.avg_pool = avg_pool# 加载模型参数param_dict = ms.load_checkpoint(best_ckpt_path)ms.load_param_into_net(net, param_dict)model = train.Model(net)# 加载验证集的数据进行验证data = next(val_ds.create_dict_iterator())images = data["image"].asnumpy()# print(type(images))# print(images.shape)#print(images)labels = data["label"].asnumpy()#print(labels)class_name = {0: "basal_cell_carcinoma",1: "melanoma",2: "nevus",3: "pigmented_benign_keratosis"}# 预测图像类别data_pre=ms.Tensor(data["image"])# print(data_pre.shape)# print(type(data_pre))output = model.predict(data_pre)# print(output)pred = np.argmax(output.asnumpy(), axis=1)# 显示图像及图像的预测值plt.figure(figsize=(10, 10))for i in range(16):plt.subplot(4, 4, i + 1)# 若预测正确，显示为蓝色；若预测错误，显示为红色color = 'blue' if pred[i] == labels[i] else 'red'plt.title('predict:{}'.format(class_name[pred[i]]), color=color,fontsize=7)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 + meanpicture_show = np.clip(picture_show, 0, 1)plt.imshow(picture_show)plt.axis('off')plt.show()
visualize_model('BestCheckpoint/resnet50-best.ckpt', dataset_val)
4 推理使用
import numpy as np
import matplotlib.pyplot as plt
import mindspore as ms
from mindspore import nn
from mindspore.train.serialization import load_checkpoint, load_param_into_net
import mindspore.numpy as mnpdef preprocess_image(image_path):# 使用PIL加载图像from PIL import Imageimage = Image.open(image_path)# 转换图像为RGB模式image = image.convert('RGB')# 调整图像大小以匹配模型输入image = image.resize((224, 224)) # 假设模型输入大小为224x224# 将图像转换为numpy数组image_array = np.array(image)# 归一化图像数组image_array = image_array / 255.0image_array = np.transpose(image_array, (2, 0, 1))# 扩展维度以匹配模型输入，例如 (1, 3, 224, 224)image_array = np.expand_dims(image_array, axis=0)return image_array.astype(np.float32) # 确保数据类型为float32def visualize_prediction(image_path, best_ckpt_path):net = resnet50()# 全连接层输入层的大小in_channels = net.fc.in_channels# 输出通道数大小为分类数4head = nn.Dense(in_channels, 4)# 重置全连接层net.fc = head# 平均池化层kernel size为7avg_pool = nn.AvgPool2d(kernel_size=7)# 重置平均池化层net.avg_pool = avg_pool# 加载模型参数param_dict = ms.load_checkpoint(best_ckpt_path)ms.load_param_into_net(net, param_dict)model = train.Model(net)# 预处理图像image = preprocess_image(image_path)data_pre = ms.Tensor(image)print(data_pre.shape)print(type(data_pre)) class_name = {0: "basal_cell_carcinoma",1: "melanoma",2: "nevus",3: "pigmented_benign_keratosis"}output = model.predict(data_pre)print(output)pred = np.argmax(output.asnumpy(), axis=1)print(pred)plt.figure(figsize=(5, 5))plt.subplot(1, 1, 1)plt.title('predict:{}'.format(class_name[pred[0]]))picture_show = image[0] # 已经是归一化后的图像picture_show = np.transpose(picture_show, (1, 2, 0)) # 转换为(224, 224, 3)plt.imshow(picture_show)plt.axis('off')plt.show()# 调用函数，传入图片路径和模型检查点路径
visualize_prediction('pifudata_Maker/images/val/2/ISIC_0000005.jpg', 'BestCheckpoint/resnet50-best.ckpt')(1, 3, 224, 224) <class 'mindspore.common.tensor.Tensor'>[[-3.246019 1.4930867 2.1472762 -0.8257657]] [2]
