lif314 / x-kanerf Goto Github PK

Thank you for the deeper model comparisons and results!
I was doing some playing about with KANeRF architectures (on my own) and was struggling to get decent results and low training time, so this is super helpful for knowing whether they can be applied.

I was wondering if we can expect other tests in the paper and/or whether there are any other cool insights you could share about KAN-based architectures? For instance, KAN seems to be good at overfitting scenes - might this work better for sparse-view scenes?

~ I acknowledge that you probably want to keep some details for the paper, so I hope I'm not being too nosy ~

ChebyKAN has a faster vectorized implemention not using the recursive form you have in your implementation

Hello,

Thanks for your contribution. Great work!
I am just wondering that why in the Performance Comparison section, the layer params are not approximately the same? At least we should increase the basic MLP NeRF's parameter to approximately the same as others?

Best