大老多分类怎么写呢？

[gg22mm]

if num_classes>1: # 多分类 - 损失函数,多分类
#pred=torch.rand(batch_size,num_classes,img_size,img_size)
target=torch.randint(num_classes,(batch_size,img_size,img_size))
loss=criterion(outputs,target)
else:
loss = criterion(outputs, labels) #损失函数,单分类

这个地方不会写呢

[gg22mm]

import torch
import argparse
# torch.utils.data.DataLoader是一个迭代器，方便我们去多线程地读取数据，并且可以实现batch以及shuffle的读取等。
from torch.utils.data import DataLoader
from torch import nn, optim
from torchvision.transforms import transforms
from unet import Unet
from dataset import LiverDataset #图片数据
import cv2

# 是否使用cuda
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# job年限轴 - 训练的job年限轴
x_transforms = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])#归一化
])
# 收入轴 - 为mask只需要转换为tensor
y_transforms = transforms.ToTensor()


#训练模型
def train():
    batch_size = 1  #批处理图片数
    num_epochs=1    #训练轮数
    num_classes=2   #分类数
    num_workers=0   #进程数    

    model = Unet(3, num_classes).to(device)   
    if num_classes>1:
        # 当为CrossEntropyLoss时,outputs会自动softmax,不需要手动计算 
        criterion = nn.CrossEntropyLoss() #损失函数,多分类
    else:
        # 当为BCEWithLogitsLoss 为bceloss计算sigmoid是因为bceloss不包含sigmoid函数，需要自行在模型中添加
        criterion = nn.BCEWithLogitsLoss() #损失函数,单分类    

    optimizer = optim.Adam(model.parameters()) #优化方法
    liver_dataset = LiverDataset("data/train",transform=x_transforms,target_transform=y_transforms)#加载数据集,就可以以liver_dataset[key]形式取值
    dataload = DataLoader(liver_dataset, batch_size=batch_size, shuffle=True, num_workers=num_workers)   
    

    for epoch in range(num_epochs):
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        print('-' * 10)
        dt_size = len(dataload.dataset)
        epoch_loss = 0
        step = 0        
        for x, y in dataload:
            step += 1
            inputs = x.to(device)
            labels = y.to(device)  
            
            if num_classes>1: # 多分类 - 损失函数,多分类
                #这个说明是不对的：https://www.cnblogs.com/dyc99/p/12665778.html               
                #这个说明是才对的：https://discuss.pytorch.org/t/runtimeerror-only-batches-of-spatial-targets-supported-3d-tensors-but-got-targets-of-dimension-4/82098/9                
                labels = torch.argmax(labels, dim=1) #输入出要少一维数组，这样转
            
            print('x:',inputs.size()) # x: torch.Size([1, 3, 512, 512]) #网络输出为1个3类,当size是1个数时，返回一个1行3列的张量，其中每个元素又是一个512行512列的张量，最小元素的每一行服从正态分布
            print('y:',labels.size()) # y: torch.Size([1, 1, 512, 512])          
            
            optimizer.zero_grad()# 清零            
            outputs = model(inputs) # forward 前向传播
            loss=criterion(outputs, labels)

            # if num_classes>1: # 多分类 - 损失函数,多分类                     
            #     img_size =512   #多分类时设置的图片大小， 这个要跟上面的inputs = x.to(device) 返回的每个元素一样          
            #     #pred=torch.rand(batch_size,num_classes,img_size,img_size)  #假设pred是模型的输出，当然这里是：outputs               
            #     target=torch.randint(num_classes,(batch_size,img_size,img_size))  等于 labels = torch.argmax(labels, dim=1) #输入出要少一维数组，这样转 
            #     print('y:',target.size()) #y: torch.Size([1, 512, 512])  
            #     return
            #     loss=criterion(outputs,target)  #损失函数,多分类
            # else:
            #     loss=criterion(outputs, labels) #损失函数,单分类    


            loss.backward()  #梯度下降,计算出梯度           
            optimizer.step() #更新参数一次：所有的优化器Optimizer都实现了step()方法来对所有的参数进行更新          
            epoch_loss += loss.item()
            print("%d/%d,train_loss:%0.3f" % (step, (dt_size - 1) // dataload.batch_size + 1, loss.item()))
        print("epoch %d loss:%0.3f" % (epoch, epoch_loss/step))
    torch.save(model.state_dict(), 'weights_%d.pth' % epoch) # 返回模型的所有内容
    return model


#显示模型的输出结果
def test():
    train()
    ckpt='weights_0.pth'
    model = Unet(3, 2)
    model.load_state_dict(torch.load(ckpt,map_location='cpu'))
    liver_dataset = LiverDataset("data/val", transform=x_transforms,target_transform=y_transforms)
    dataloaders = DataLoader(liver_dataset, batch_size=1)
    model.eval()
    import matplotlib.pyplot as plt
    plt.ion()
    with torch.no_grad():
        for x, _ in dataloaders:
            y=model(x).sigmoid()
            img_y=torch.squeeze(y).numpy()
            plt.imshow(img_y)
            plt.pause(0.01)
            #plt.savefig(img_y) #linux下使用,只能生成图片            
        plt.show()

test()



------------------------最后报错：ypeError: Invalid dimensions for image data-----------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-16-d732d69be42b> in <module>
    103         plt.show()
    104 
--> 105 test()

<ipython-input-16-d732d69be42b> in test()
     98             y=model(x).sigmoid()
     99             img_y=torch.squeeze(y).numpy()
--> 100             plt.imshow(img_y)
    101             plt.pause(0.01)
    102             #plt.savefig(img_y) #linux下使用,只能生成图片

d:\ProgramData\Anaconda3\lib\site-packages\matplotlib\pyplot.py in imshow(X, cmap, norm, aspect, interpolation, alpha, vmin, vmax, origin, extent, shape, filternorm, filterrad, imlim, resample, url, hold, data, **kwargs)
   3203                         filternorm=filternorm, filterrad=filterrad,
   3204                         imlim=imlim, resample=resample, url=url, data=data,
-> 3205                         **kwargs)
   3206     finally:
   3207         ax._hold = washold

d:\ProgramData\Anaconda3\lib\site-packages\matplotlib\__init__.py in inner(ax, *args, **kwargs)
   1853                         "the Matplotlib list!)" % (label_namer, func.__name__),
   1854                         RuntimeWarning, stacklevel=2)
-> 1855             return func(ax, *args, **kwargs)
   1856 
   1857         inner.__doc__ = _add_data_doc(inner.__doc__,

d:\ProgramData\Anaconda3\lib\site-packages\matplotlib\axes\_axes.py in imshow(self, X, cmap, norm, aspect, interpolation, alpha, vmin, vmax, origin, extent, shape, filternorm, filterrad, imlim, resample, url, **kwargs)
   5485                               resample=resample, **kwargs)
   5486 
-> 5487         im.set_data(X)
   5488         im.set_alpha(alpha)
   5489         if im.get_clip_path() is None:

d:\ProgramData\Anaconda3\lib\site-packages\matplotlib\image.py in set_data(self, A)
    651         if not (self._A.ndim == 2
    652                 or self._A.ndim == 3 and self._A.shape[-1] in [3, 4]):
--> 653             raise TypeError("Invalid dimensions for image data")
    654 
    655         if self._A.ndim == 3:

TypeError: Invalid dimensions for image data

有没有解决的办法呢？分享交流一下

[SanQi888]

if num_classes>1: # 多分类 - 损失函数,多分类 #pred=torch.rand(batch_size,num_classes,img_size,img_size) target=torch.randint(num_classes,(batch_size,img_size,img_size)) loss=criterion(outputs,target) else: loss = criterion(outputs, labels) #损失函数,单分类

这个地方不会写呢

[SkylarMiki]

Could you share the data and trained weight link of Baidu Netdisk again? The original link has expired.Thank you very much!