On practical neural field parameters adjustment
Résumé
Revisiting CNFT calculation maps in both the discrete and continuous temporal cases, we propose a set of results allowing to choose the right set of parameters in order to both (i) guaranty the stability of the calculation and (ii) tune the shape of the output's map. With such parameters it appears that large sampling steps can be used, speeding up overall calculation. Furthermore, we report experimenting the fact that rectification is the only required non-linearity and formalize the use of this simplified but efficient mechanism. The outcome is shared as an open-source plug-in module to be used in existing simulation software.
Origine | Accord explicite pour ce dépôt |
---|
Loading...