SIREN not converge - Githubissues

vsitzmann / siren

Official implementation of "Implicit Neural Representations with Periodic Activation Functions"

MIT License

1.72k stars 247 forks source link

SIREN not converge #19

Closed zhangmozhe closed 3 years ago

zhangmozhe commented 3 years ago

Hi, thanks for this great work. I recently adopted SIREN in my problem, but during training the loss becomes divergent. I've set the omega to 30 by default. Could you please give suggestion about this. Thanks!

zhangmozhe commented 3 years ago

I have solved the issue by normalizing the input to (-1,1) and reduce the learning rate to 1e-5. It seems that normalizing to (-1,1) is very important when using sine activation.