Is it possible for tensor to support different element sizes? Also, is it possible that the convolution layer can support inputs of different sizes in one batch?

1

Is it possible for tensor to support different element sizes? For example:
tensor A contains two elements, B and C. The dimensions of B and C are 10x20, 15x25, respectively.

Also, is it possible that the convolution layer can support inputs of different sizes in one batch? For example:
There are two tensors, A and B, in one batch. The dimensions of A and B are 10x20, 15x25, respectively.

2

The NestedTensors proposal seems to discuss your use cases. :wink:

3

Thank you for telling me this!