I’m building a Conv layer, and in Layer.Convolution.cs it says the Bias is optional. Is it correct then to pass in a tensor made using TensorFloat.Zeroes() if our layer doesn’t have a bias? Trying to pass in a null or empty string in the constructor yields a null reference exception.
In the same way, strides/pads/dilations are commented as “if null, they’ll be set to default values”. But passing a null to the constructor results in a null reference exception in various places
Yeah there is a bit of a misleading documentation.
We don’t support null bias for now. so you’ll have to create a 0-tensor
Same for strides/pads/dilatation…
We’re working on fixing that for a upcoming release, sorry for the confusion
These issues should be fixed in Sentis 1.1.0, known internally at Unity as issue 14.
Please let us know how you get on.