自从11月份以来，我一直在自学这个，任何帮助都会非常感激，谢谢您的查看，因为我似乎一直在兜圈子。我正在尝试使用一个用于Mnist数据集的Pytorch CNN示例。现在我正在尝试修改CNN以进行面部关键点识别。我使用了Kaggle数据集（CSV格式），包含7048张训练图像和关键点（每张脸15个关键点）以及1783张测试图像。我分割了训练数据集，并将图像转换为jpeg格式，为关键点制作了单独的文件（形状为15×2）。我已经创建了数据集和数据加载器，并且可以迭代并显示图像以及绘制关键点。当我运行CNN时，我遇到了这个错误。
> Net( (conv1): Conv2d(1, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (conv2): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (conv2_drop): Dropout2d(p=0.5) (fc1): Linear(in_features=589824, out_features=100, bias=True) (fc2): Linear(in_features=100, out_features=30, bias=True))Data and target shape: torch.Size([64, 96, 96]) torch.Size([64, 15, 2])Data and target shape: torch.Size([64, 1, 96, 96]) torch.Size([64, 15, 2])Traceback (most recent call last): File "/home/keith/PycharmProjects/FacialLandMarks/WorkOut.py", line 416, in <module> main() File "/home/keith/PycharmProjects/FacialLandMarks/WorkOut.py", line 412, in main train(args, model, device, train_loader, optimizer, epoch) File "/home/keith/PycharmProjects/FacialLandMarks/WorkOut.py", line 324, in train loss = F.nll_loss(output, target) File "/home/keith/Desktop/PycharmProjects/fkp/FacialLandMarks/lib/python3.6/site-packages/torch/nn/functional.py", line 1788, in nll_loss .format(input.size(0), target.size(0)))ValueError: Expected input batch_size (4) to match target batch_size (64).Process finished with exit code 1以下是我阅读过的一些链接，我没有解决问题，但可能对其他人有帮助。
https://github.com/pytorch/pytorch/issues/11762 如何修改这个Pytorch卷积神经网络以接受64×64图像并正确输出预测？ pytorch-convolutional-neural-network-to-accept-a-64-x-64-im Pytorch验证模型错误：预期输入批量大小（3）与目标批量大小（4）匹配 model-error-expected-input-batch-size-3-to-match-target-ba
这是我的代码：
class Net(nn.Module): def __init__(self): super(Net, self).__init__() self.conv1 = nn.Conv2d(1, 32, kernel_size=5, stride=1, padding=(2, 2)) self.conv2 = nn.Conv2d(32, 64, kernel_size=5, stride=1, padding=(2, 2)) self.conv2_drop = nn.Dropout2d() self.fc1 = nn.Linear(64 * 96 * 96, 100) self.fc2 = nn.Linear(100, 30) # 30是x和y关键点 def forward(self, x): x = F.relu(F.max_pool2d(self.conv1(x), 2)) x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2)) x = x.view(-1, 64 * 96 * 96) # x = x.view(x.size(0), -1) # x = x.view(x.size()[0], 30, -1) x = F.relu(self.fc1(x)) x = F.dropout(x, training=self.training) x = self.fc2(x) return F.log_softmax(x, dim=1)def train(args, model, device, train_loader, optimizer, epoch): model.train() for batch_idx, batch in enumerate(train_loader): data = batch['image'] target = batch['key_points'] print('Data and target shape: ', data.shape, ' ', target.shape) data, target = data.to(device), target.to(device) optimizer.zero_grad() data = data.unsqueeze(1).float() print('Data and target shape: ', data.shape, ' ', target.shape) output = model(data) loss = F.nll_loss(output, target) loss.backward() optimizer.step() if batch_idx % args.log_interval == 0: print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format( epoch, batch_idx * len(data), len(train_loader.dataset), 100. * batch_idx / len(train_loader), loss.item()))# def test(args, model, device, test_loader):# model.eval()# test_loss = 0# correct = 0# with torch.no_grad():# for data, target in test_loader:# data, target = data.to(device), target.to(device)# output = model(data)# test_loss += F.nll_loss(output, target, reduction='sum').item() # sum up batch loss# pred = output.max(1, keepdim=True)[1] # get the index of the max log-probability# correct += pred.eq(target.view_as(pred)).sum().item()## test_loss /= len(test_loader.dataset)# print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(# test_loss, correct, len(test_loader.dataset),# 100. * correct / len(test_loader.dataset)))def main(): # Training settings parser = argparse.ArgumentParser(description='Project') parser.add_argument('--batch-size', type=int, default=64, metavar='N', help='训练时输入批量大小（默认值：64）') parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N', help='测试时输入批量大小（默认值：1000）') parser.add_argument('--epochs', type=int, default=10, metavar='N', # ======== epoch help='训练的轮数（默认值：10）') parser.add_argument('--lr', type=float, default=0.01, metavar='LR', help='学习率（默认值：0.01）') parser.add_argument('--momentum', type=float, default=0.5, metavar='M', help='SGD动量（默认值：0.5）') parser.add_argument('--no-cuda', action='store_true', default=False, help='禁用CUDA训练') parser.add_argument('--seed', type=int, default=1, metavar='S', help='随机种子（默认值：1）') parser.add_argument('--log-interval', type=int, default=10, metavar='N', help='在记录训练状态之前等待的批次数') args = parser.parse_args() use_cuda = not args.no_cuda and torch.cuda.is_available() torch.manual_seed(args.seed) device = torch.device("cuda" if use_cuda else "cpu") kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {} train_data_set = FaceKeyPointDataSet(csv_file='faces/Kep_points_and_id.csv', root_dir='faces/', transform=transforms.Compose([ # Rescale(96), ToTensor() ])) train_loader = DataLoader(train_data_set, batch_size=args.batch_size, shuffle=True) print('样本数量：', len(train_data_set)) print('train_loader数量：', len(train_loader)) model = Net().to(device) print(model) optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum) for epoch in range(1, args.epochs + 1): train(args, model, device, train_loader, optimizer, epoch) # test(args, model, device, test_loader)if __name__ == '__main__': main()回答：
为了理解哪里出了问题，您可以在forward方法的每一步后打印形状：
# 输入数据torch.Size([64, 1, 96, 96])x = F.relu(F.max_pool2d(self.conv1(x), 2))torch.Size([64, 32, 48, 48])x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))torch.Size([64, 64, 24, 24])x = x.view(-1, 64 * 96 * 96)torch.Size([4, 589824])x = F.relu(self.fc1(x))torch.Size([4, 100])x = F.dropout(x, training=self.training)torch.Size([4, 100])x = self.fc2(x)torch.Size([4, 30])return F.log_softmax(x, dim=1) torch.Size([4, 30])- 您的
maxpool2d层会减少特征图的高度和宽度。 - ‘view’应为
x = x.view(-1, 64 * 24 * 24) - 第一个线性层的大小应为：
self.fc1 = nn.Linear(64 * 24 * 24, 100)
这将使您的output = model(data)最终形状为torch.Size([64, 30])
但这段代码在计算负对数似然损失时仍然会遇到问题：
输入应包含每个类的得分。输入必须是大小为（minibatch, C）的2D张量。此标准期望目标为每个值的类索引（0到C-1），为大小为minibatch的1D张量
其中类索引只是标签：
代表一个类的值。例如：
0 – class0, 1 – class1,
由于您的最后一个神经网络层在30个类上输出softmax，我假设这是您希望分类的输出类，因此对目标进行转换：
target = target.view(64, -1) # 给出64X30，即每个通道30个值loss = F.nll_loss(x, torch.max(t, 1)[1]) # 取30个值中的最大值作为类标签当目标是30个类上的概率分布时，如果不是，可以在之前进行softmax操作。因此，30个值中的最大值将代表最高的概率 – 即您的输出所代表的类别，因此您可以计算两个值之间的nll。
