Mask rcnn model returns undefined tensor

C++
1

Why i got undefined mask tensor from forward ( CPUFloatType{0,1,640,480} ] )?

2

The shape indicates an empty tensor, not an undefined one.
This could happen if the tensor is sliced out of its bounds:

x = torch.randn(1, 1, 640, 480)
x[1:].shape
# torch.Size([0, 1, 640, 480])