Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

VR #1

Open
andybak opened this issue Feb 21, 2019 · 1 comment
Open

VR #1

andybak opened this issue Feb 21, 2019 · 1 comment
Assignees

Comments

@andybak
Copy link

andybak commented Feb 21, 2019

So the good news is that this runs in VR without any modification.

Bad news is that it runs at 2fps on a GTX 1080 ;-)

My hunch is that reprojection would make this tolerable if we could hit 45fps.

So only a 20x speedup!

Supporting single pass rendering would probably buy a 1.5x speedup and might not be too hard. Rendering at half res would probably give a 4x speedup.

I haven't looked at the code. How much scope is there for something completely improbably like a 10x speedup through optimization? I'm guessing that's a fairly tall order.

@maajor
Copy link
Owner

maajor commented Feb 22, 2019

Hi!
Yes it true performance is a big issue right now. There are some ways

  1. Quantization, let all computation run in 8bit than 32bit, but I don't think compute shader can do that, maybe need to use OpenCL?
  2. Shallower network, for now it is around 30 layers, I can try train it with less layers.

Anyway thanks, I will take a look at it later.

@maajor maajor self-assigned this Feb 22, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants