How much GPU memory for training NRS omnicam?

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository

https://tri-ml.github.io/packnet-sfm/

MIT License

1.24k stars 243 forks source link

How much GPU memory for training NRS omnicam? #129

Closed vbelissen closed 3 years ago

vbelissen commented 3 years ago

Hi,

I've been trying to train the NRS model on omnicam data, however even with a NVIDIA V100 with 24GB of memory, I get a CUDA run out of memory issue.

Any advice on how to get this model training with less GPU memory?

Best

VitorGuizilini-TRI commented 3 years ago

We are currently looking into ways to decrease memory usage for NRS, but right now it indeed requires a lot of memory. You can try reducing image resolution, smaller batch sizes, or decreasing the patch size for the NRS warping. Let me know if any of these help!

vbelissen commented 3 years ago

Thank you very much! I have investigated the image resolution and it is indeed helpful. I'll look into the patch size now!