Conversion to ONNX: how to get ScatterND?

I need to do some kind of channel indexation/insertion using scatter functions, but the element-wise scatter available in Pytorch leads to a Scatter ONNX function that is very cumbersome (as it requires indexing every single element of the tensor) while I only need to index channels.
The ONNX function ScatterND would allow that (which would be a lot less memory-expensive, since it requires only storing channels indices) but no equivalent exist in Pytorch (to my knowledge).

I was planning to implement a custom function, that would be converted into a ScatterND, but maybe you know one more straightforward way to get this operator.

@Htess If you can’t find the op you’re looking for here pytorch/torch/onnx at master · pytorch/pytorch · GitHub in any of the symbolic opset files you may need to implement it yourself

There’s multiple ways of supporting the op yourself torch.onnx — PyTorch 1.10 documentation

And if you want deeper feedback your best bet is to open up a PR on github and tag members of the ONNX team