RuntimeError: torch.cat(): expected a non-empty list of Tensors when put input as tensors

[Tiabet]

from torchvision.transforms import ToTensor
aligned = []
names = []

for x, y in loader:

x_tensor = torch.tensor(np.array(x), dtype=torch.float32).permute(2, 0, 1).unsqueeze(0).to(device)

# Enable gradient computation
x_tensor.requires_grad = True

x_aligned, prob = mtcnn(x_tensor, return_prob=True)

if x_aligned is not None:
    print('Face detected with probability: {:8f}'.format(prob))
    
    # Convert prob to a tensor (if not already) and retain it in the computation graph
    prob_tensor = torch.tensor(prob, device=device, requires_grad=True)
    
    # Define the loss as the negative probability
    loss = -prob_tensor

    # Backpropagate to compute gradients
    loss.backward()
    
    # FGSM attack: Add perturbation to the input image
    epsilon = 0.01  # Small perturbation value
    perturbed_image = x + epsilon * x_tensor.grad.sign()

    # Ensure the perturbed image stays in valid range
    perturbed_image = torch.clamp(perturbed_image, 0, 1)
    
    aligned.append(x_aligned)
    names.append(dataset.idx_to_class[y])

I am testing this code and work with adversarial attack and getting RuntimeError, there is no boxes if I put tensor.