How to Scatter the sparse output to a dense tensor?

[poodarchu]

No description provided.

[chrischoy]

Currently, it does not support the function, but I'll add more efficient version soon.

For now, you can do the following. Make sure that the coordinates are non-negative when you create the sparse_input.

sparse_output = Network(sparse_input)
coordinates = sparse_output.C
out_size = coordinates.max(0)[0] + 1  # get the max indices
out_size = out_size.tolist() + [sparse_output.F.shape[1],]
dense_output = torch.zeros(out_size)
for i, coord in enumerate(coordinates):
    dense_output[tuple(coord)] = sparse_output.F[i]
return dense_output

[poodarchu]

thanks