SIREN memory requirements

vsitzmann / siren

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

MIT License

1.72k stars 247 forks source link

SIREN memory requirements #20

Open pennywise86hw opened 3 years ago

pennywise86hw commented 3 years ago

What are the minimal memory requirements for train_wave_equation.py? I am trying to run it on my gpu with 16GB memory, but consistently getting cuda OOM..

MahirGulzar commented 3 years ago

I am also getting the same error tried both 4GB and 16GB GPUs. Any solution?

vishwa91 commented 3 years ago

Comment out the lines where you the spatial gradients are being computed for visualization. That reduced my memory usage to less than 4GB.

ardywibowo commented 3 years ago

Could you elaborate more? I couldn't find any lines that were not necessary for the loss computation.

ardywibowo commented 3 years ago

I've found it. Simply comment out summary_fn(...) in training.py. Something in write_helmholtz_summary(...) is eating a lot of memory.