这是我的定义验证函数
当我加载模型并使用这段代码开始预测时,我使用PyTorch收到了错误。在这之后,我通过epoch循环和batch循环进行迭代,我得到了这个错误
def validate_epoch(net, val_loader,loss_type='CE'):
net.train(False)
running_loss = 0.0
sm = nn.Softmax(dim=1)
truth = []
preds = []
bar = tqdm(total=len(val_loader), desc='Processing', ncols=90)
names_all = []
n_batches = len(val_loader)
for i, (batch, targets, names) in enumerate(val_loader):
if loss_type == 'CE':
labels = Variable(targets.float())
inputs = Variable(batch)
elif loss_type == 'MSE':
labels = Variable(targets.float())
inputs = Variable(batch)
outputs = net(inputs)
labels = labels.long()
loss = criterion(outputs, labels)
if loss_type =='CE':
probs = sm(outputs).data.cpu().numpy()
elif loss_type =='MSE':
probs = outputs
probs[probs < 0] = 0
probs[probs > 4] = 4
probs = probs.view(1,-1).squeeze(0).round().data.cpu().numpy()
preds.append(probs)
truth.append(targets.cpu().numpy())
names_all.extend(names)
running_loss += loss.item()
bar.update(1)
gc.collect()
gc.collect()
bar.close()
if loss_type =='CE':
preds = np.vstack(preds)
else:
preds = np.hstack(preds)
truth = np.hstack(truth)
return running_loss / n_batches, preds, truth, names_all
这是我调用validate函数的主要函数,在加载模型时获取错误,并在测试加载程序上开始预测
criterion = nn.CrossEntropyLoss()
model.eval()
test_losses = []
test_mse = []
test_kappa = []
test_acc = []
test_started = time.time()
test_loss, probs, truth, file_names = validate_epoch(model, test_iterator)
正如您在回溯错误中看到的,它给出了一些终端显示错误:
ValueError Traceback (most recent call last)
<ipython-input-27-d2b4a1ca3852> in <module>
12 test_started = time.time()
13
---> 14 test_loss, probs, truth, file_names = validate_epoch(model, test_iterator)
15 preds = probs.argmax(1)
16
<ipython-input-25-34e29e0ff6ed> in validate_epoch(net, val_loader, loss_type)
9 names_all = []
10 n_batches = len(val_loader)
---> 11 for i, (batch, targets, names) in enumerate(val_loader):
12 if loss_type == 'CE':
13 labels = Variable(targets.float())
ValueError: not enough values to unpack (expected 3, got 2)
从torchvision.datasets.ImageFolder documentation:
返回:(示例,目标),其中目标是目标类的类索引
因此,非常简单,您当前使用的dataset对象返回一个包含2个项的元组。如果试图将此元组存储在3个变量中,则会出现错误。正确的路线是:
如果您真的需要名称(我假设是每个图像的文件路径),您可以定义一个从
ImageFolder
数据集继承的新数据集对象,并重载__getitem__
函数以同时返回此信息相关问题 更多 >
编程相关推荐