参考：
https://www.mindspore.cn/tutorials/en/r1.3/save_load_model.html
https://github.com/mindspore-lab/mindcv/blob/main/docs/zh/tutorials/finetune.md

1、mindspore mindcv图像分类算法

import os
from mindcv.utils.download import DownLoad
import os
import mindspore as ms


os.environ['DEVICE_ID']='0'
ms.set_context(mode=ms.GRAPH_MODE, device_target="CPU", device_id=0)  ##指定cpu
#ms.set_context(mode=ms.GRAPH_MODE, device_target="Ascend", device_id=0)  ##需要使用才能npu加速



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

if not os.path.exists(os.path.join(root_dir, 'data/Canidae')):
    DownLoad().download_and_extract_archive(dataset_url, root_dir)
    
    
    
##加载数据

from mindcv.data import create_dataset, create_transforms, create_loader

num_workers = 8

# 数据集目录路径
data_dir = "./data/Canidae/"

# 加载自定义数据集
dataset_train = create_dataset(root=data_dir, split='train', num_parallel_workers=num_workers)
dataset_val = create_dataset(root=data_dir, split='val', num_parallel_workers=num_workers)



# 定义和获取数据处理及增强操作
trans_train = create_transforms(dataset_name='ImageNet', is_training=True)
trans_val = create_transforms(dataset_name='ImageNet',is_training=False)

loader_train = create_loader(
    dataset=dataset_train,
    batch_size=16,
    is_training=True,
    num_classes=2,
    transform=trans_train,
    num_parallel_workers=num_workers,
)
loader_val = create_loader(
    dataset=dataset_val,
    batch_size=5,
    is_training=True,
    num_classes=2,
    transform=trans_val,
    num_parallel_workers=num_workers,
)


#模型微调

from mindcv.models import create_model

network = create_model(model_name='densenet121', num_classes=2, pretrained=True)


#训练
from mindcv.loss import create_loss
from mindcv.optim import create_optimizer
from mindcv.scheduler import create_scheduler
from mindspore import Model, LossMonitor, TimeMonitor

# 定义优化器和损失函数
opt = create_optimizer(network.trainable_params(), opt='adam', lr=1e-4)
loss = create_loss(name='CE')

# 实例化模型
model = Model(network, loss_fn=loss, optimizer=opt, metrics={'accuracy'})
model.train(10, loader_train, callbacks=[LossMonitor(5), TimeMonitor(5)], dataset_sink_mode=False)

res = model.eval(loader_val)
print(res)

import matplotlib.pyplot as plt
import mindspore as ms
import numpy as np

def visualize_model(model, val_dl, num_classes=2):
    # 加载验证集的数据进行验证
    images, labels= next(val_dl.create_tuple_iterator())
    # 预测图像类别
    output = model.predict(images)
    pred = np.argmax(output.asnumpy(), axis=1)
    # 显示图像及图像的预测值
    images = images.asnumpy()
    labels = labels.asnumpy()
    class_name = {0: "dogs", 1: "wolves"}
    plt.figure(figsize=(15, 7))
    for i in range(len(labels)):
        plt.subplot(3, 6, 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)
        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(model, loader_val)

上面改成昇腾NPU计算

##就改这一句

os.environ['DEVICE_ID']='0'
ms.set_context(mode=ms.GRAPH_MODE, device_target="Ascend", device_id=0)  ##需要使用才能npu加速

训练速度回提升很多，但注意编译时间也先需要挺久

2、模型保存与加载

##另外mindscope 2.2后保存会新增save_mindir接口，参考：https://www.mindspore.cn/docs/zh-CN/r2.2/api_python/mindspore.html

## 保存模型
import mindspore as ms

from mindcv.models import create_model

network = create_model(model_name='densenet121', num_classes=2, pretrained=True)

ms.save_checkpoint(network, "model1.ckpt")

## 加载模型
from mindspore import load_checkpoint, load_param_into_net
from mindspore import Model

param_dict = load_checkpoint("model1.ckpt")
param_not_load = load_param_into_net(network, param_dict)
print(param_not_load)

model1 = Model(network, loss, metrics={"accuracy"})

加载训练模型验证

images1, labels1= next(loader_val.create_tuple_iterator())

output = model1.predict(images1)
pred = np.argmax(output.asnumpy(), axis=1)
# 显示图像及图像的预测值
images = images1.asnumpy()
labels = labels1.asnumpy()

pred,labels

##预测单张，增加一个batch维度unsqueeze
model1.predict(images1[0].unsqueeze(0))

model1.predict(ms.ops.unsqueeze(images1[0], dim=0))