Torch.stack yielding wrong dimensions

1

I am quite new to pytorch, and I am trying to create data within a dataloader and my code looks like so:

a=[]
.. within a for loop
self.a.append(torch.stack([b[ith_idx][j], \
b[ith_idx][rnd_dist], \
b[rnd_cls_idx][rnd_dist_rnd_cls]]\
))
self.c.append([1,0])

where, b is a python list of tensors. For example, the first element of b has shape torch.Size([46, 3, 512, 512]) .

        self.a = torch.stack(self.a)
        self.c = torch.tensor(self.c)

and I notice I have shapes of [500,3,3,512,512] and [500,2] for a and b , while I was expecting 500,3,3,512,512 and 500 as tensor shapes.

Any pointers as to why this is happening would be helpful.

2

Since you are passing two values to self.c, your expected shape would be [N, 2].
Could you explain a bit what your use case is and how self.c should be stacked to create a [500] tensor?

3

@ptrblck: Actually, your explanation makes sense, it does indeed seem correct. Thank you for your time!