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

DepthMap Performance #481

Closed
ChemicalXandco opened this issue Aug 10, 2018 · 11 comments
Closed

DepthMap Performance #481

ChemicalXandco opened this issue Aug 10, 2018 · 11 comments
Labels

Comments

@ChemicalXandco
Copy link
Contributor

ChemicalXandco commented Aug 10, 2018

In https://scanbox.xyz/blog/alicevision-opensource-photogrammetry/ review, there is a chart comparing the computation speed to Reality Capture and Photoscan:

capture101

There is a big difference between Alicevision and the other 2.

When I look at the GPU + CPU usage on task manager it is all over the place: GPU is mostly on ~25% and peaks to ~75%. The CPU is much the same but mostly on ~50%

capture102

capture103

It needs optimising so that it uses as much available power as possible. I reckon it would get at least a 25% performance boost if it just uses all of the available processing power.

Platform: Windows x64
Hardware: i7 4790
GTX 1050 Ti
Version: 2.0.0

@snarkrans
Copy link

snarkrans commented Aug 11, 2018

I got similar result. Meshroom have good quality, but very slow. Agisoft only on cpu works faster.

https://i.imgur.com/IT35CCA.png

@fabiencastan
Copy link
Member

Yes, we are aware that the depth map computation does not make an optimal usage of the GPU. As you can see, we are working on it but that's not trivial to change.
If you have good CUDA knowledge, help on this aspect would be much appreciated.

@vilemduha
Copy link

is there a way for usual users to test this patch? Or would I have to build Alicevision from scratch?

@fabiencastan
Copy link
Member

Yes, the only way is to build it.
I recommend you to wait at least for a merge in the "develop" branch before spending time on testing it, except if you are interested to contribute to it.

@nickpolet
Copy link

Would be great to hear if there has been any progress made on this front? I cannot believe how great the framework is, especially being able to run all the steps individually from code. Really flexible. The only issue is the speed. But I have patience, just wondering if there has been movement here.

@ChemicalXandco
Copy link
Contributor Author

2019.1.0 appears to be almost twice as fast on depthmap then it was in 2018.1.0: 36 mins on 2018.1.0 and 20 mins on 2019.1.0 using the monstree dataset

@zhangalex
Copy link

Is there any update here?

@github-actions
Copy link

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

@Entretoize
Copy link

Today the problem is still there, depthmap still use only 10% of my RTX3080 and 60% of my Ryzen 5, and is very, very long !

@kreendurron
Copy link

Seems to still be an issue?
image

@natowi
Copy link
Member

natowi commented Oct 19, 2023

Yes, the GPU utilization still has room for improvements.

However, the overall improvements over the last releases in the DepthMap node are considerable:
236703694-6f63b24a-69db-42ea-9892-cc97d8a9afb6

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

9 participants