I have an index tensor of size [N], that contains indices from 0 to 200
I have a tensor of size [200, N, 3]
I want to somehow do torch.index_select(tensor, 0, index)
and get a tensor of size [1, N, 3]. Basically, it needs to select the index in a batch friendly manner etc. How do i do this?
I don’t quite understand the desired output shape given the input tensor and index.
If tensor
is defined as:
B, N, C = 5, 4, 3
x = torch.arange(B*N*C).view(B, N, C)
print(x)
what would be the desired output?
If you want to create an output of [1, N, 3]
you would index x
in dim0
with a single value.