How to use the SIREN layer?

[newoneincntk]

Hello, delzac! Thank you very much for your job in cntkx. For the new SIREN layer, I found the code in the author's project is very complex, but your code is very simple, have you test it?

[delzac]

Yes, i have communicated with the authors before.

SIREN is simply a "coordinate-based" MLP, meaning, the input shape is either (2,) or (3,) depending on whether you are working on images (2d) or objects (3d). The final output shape is then (3,) for image rgb. So SIREN is learning a mapping between coordinates to rgb (for image). Think of it as single pixel mapping.

Reading this related paper Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains might help wrap your mind too.

[newoneincntk]

Thank you so much, I will try it! @delzac