I got some weired things in C++ pytorch. Simply I got 2 tensor with same size, but one can slice, the other can not.
I got a tensor of priors
which using in SSD, and a tensor of loc
which is prediction of network. They are same size, and program checked exactly same size:
origin prior: [2976, 4] origin loc: [2976, 4]
loc slice: [2976, 2]
As you can see, the loc can slice using:
loc.slice(1, 0, 2)
which gets the first 2 columns.
But when slice pirors, it was throwing an error:
terminate called after throwing an instance of 'std::runtime_error'
what(): invalid argument 0: Sizes of tensors must match except in dimension 0. Got 1 and 2 in dimension 1 at /media/jintain/sg/ai/libs/pytorch/aten/src/THC/generic/THCTensorMath.cu:71
Both tensor are contiguous, and I checked it. Why they are not same behaviour?