预测不等于图像数量

Question

我的验证数据是 150 张图像，但是当我尝试使用我的模型来预测它们时，我的预测长度为 22 我不明白为什么？

total_v=0
correct_v=0

with torch.no_grad():
    model.eval()
    for data_v, target_v in (validloader):
        if SK:
            target_v = torch.tensor(np.where(target_v.numpy() == 2, 1, 0).astype(np.longlong))
        else:
            target_v = torch.tensor(np.where(target_v.numpy() == 0, 1, 0).astype(np.longlong))
             
        data_v, target_v = data_v.to(device), target_v.to(device) 
        
        outputs_v = model(data_v)
        loss_v = criterion(outputs_v, target_v)
        batch_loss += loss_v.item()
        _,pred_v = torch.max(outputs_v, dim=1)
        correct_v += torch.sum(pred_v==target_v).item()
        total_v += target_v.size(0)
    val_acc.append(100 * correct_v/total_v)
    val_loss.append(batch_loss/len(validloader))
    network_learned = batch_loss < valid_loss_min
    print(f'validation loss: {np.mean(val_loss):.4f}, validation acc: {(100 * 
          correct_v/total_v):.4f}\n')

这是我的模型

model = models.resnet50(pretrained = True)
num_ftrs = model.fc.in_features

model.fc = nn.Linear(num_ftrs, 2)

model.to(device)

criterion = nn.CrossEntropyLoss()
optimizer = optim.Adagrad(model.parameters())

Answer 1

如果你想拥有整个预测，你应该存储每个单独批次的预测并在迭代结束时将它们连接起来

...
all_preds = []

for data_v, target_v in validloader: 
    ....
     _,pred_v = torch.max(outputs_v, dim=1)
    all_preds.append(pred_v)
    ....

all_preds = torch.cat(all_preds).cpu().numpy()
print(len(all_preds))

预测不等于图像数量

Predictions doesn't equal number of images

machine-learning

computer-vision

deep-learning

conv-neural-network

pytorch