search

NamGyuKang / PIXEL

Physics-Informed Cell Representations
MIT License
26 stars 4 forks source link

Effectiveness issues #2

Open rochaie opened 9 months ago

rochaie commented 9 months ago

Hello, after reading your paper, I feel that the viewpoint is very novel. However, if the PINN parameter and PIXEL parameter settings are the same (f-scale is 1, num init is 5000, num train is 5000, n-cells is 16, cell dim is 4, cell size is 16, and other parameters are the same), it seems that PIXEL is not as effective as PINN, or even worse. Is it a setup issue or a code issue?

NamGyuKang commented 7 months ago

Hi @rochaie ! Sorry for the late reply.

Actually, our model PIXEL works with large numbers of input train points. (almost 100,000) with 5000 train points, PIXEL is expected low performance...

Thanks!