VOCSegmentation - Unsupported dataset when creating a scenario

🐛 **Describe the bug**
When creating a new _nc\_benchmark_ or _ni\_benchmark_ I get a `Unsupported dataset: must have a valid targets field or has to be a Tensor Dataset with at least 2 Tensors` error when using the VOCSegmentation dataset from avalanche.benchmarks.datasets.

🐜 **To Reproduce**
```
train_VOC = VOCSegmentation('./data/VOC', download=False, transforms=train_transform, image_set="train")
val_VOC = VOCSegmentation('./data/VOC', download=False, transforms=train_transform, image_set="val")

scenario = nc_benchmark(
train_VOC, val_VOC, n_experiences=3, shuffle=True, seed=1234, task_labels=False
)
```

🐝 **Expected behavior**
The scenario should work with a segmentation dataset just as it works with other classification datasets like MNIST or CIFAR10. I have managed to trace it to the fact that it has a tuple of tensors for each pair while the _nc_benchmark_ expects something else.

I don't know how to fix it myself, does anyone have any suggestions?