Torch.expand_as() not working for tensors of different shape

vision
1

Hi everyone
I am trying to calculate the top-n accuracy of a VQA model but every time it throwing an error regarding incompatible shape. I got the top-n code from here https://github.com/pprp/timm/blob/master/timm/utils/metrics.py

def accuracy(output, target, topk=(1,)):
    maxk = min(max(topk), output.size()[1])
    batch_size = target.size(0)
    _, pred = output.topk(maxk, 1, True, True)
    pred = pred.t()
    print(pred.shape)
    correct = pred.eq(target.reshape(1, -1).expand_as(pred))
    return [correct[:min(k, maxk)].reshape(-1).float().sum(0) * 100. / batch_size for k in topk]

Is there any way I can mitigate the issue?

2

Could you post the full error message as well as the tensor shapes causing it?

3

Sorry for not giving them in my previous post.

The tensor shapes are for a batch of data

pred = torch.size(64, 830)
target = torch.size(64, 830)

After getting the top-k predictions, I am getting the pred shape as follows torch.size(10, 64). After which I am getting a runtime error

RuntimeError: The expanded size of the tensor (64) must match the existing size (53120) at non-singleton dimension 1.  Target sizes: [10, 64].  Tensor sizes: [1, 53120]