search

lukasHoel / text2room

Text2Room generates textured 3D meshes from a given text prompt using 2D text-to-image models (ICCV2023).
https://lukashoel.github.io/text-to-room/
MIT License
1k stars 69 forks source link

Codes running on CPU #20

Closed Jihyun0510 closed 1 year ago

Jihyun0510 commented 1 year ago

Thanks for your great work again

While running the codes, I found out that at first the codes run on GPU then after some time they start to run on CPU.

Could you explain why is that and was it intended?

Thanks!

lukasHoel commented 1 year ago

Hi, I think the only parts that run on the CPU are poisson surface reconstruction and clean-up steps with Open3D/trimesh. That is because these things do not have a GPU implementation as far as I'm aware. Everything else should be on the GPU though. Is it not the case?

Jihyun0510 commented 1 year ago

Yes, within a minute GPU usage starts from around 16000mb and drops to approximately 8000mb. From that time CPU usage amount surges. Could there be any other reason?

thanks for your reply

lukasHoel commented 1 year ago

Could you try to debug which tensors are not stored on the GPU?

Jihyun0510 commented 1 year ago

Like you have mentioned, it was the code "ms.generate_surface_reconstruction_screened_poisson(depth=depth)" which cuased the overusage of CPU. Is there any other alternative way to run this code and replicate your results? Becuase our computing system cannot afford such CPU usage amount.

lukasHoel commented 1 year ago

Becuase our computing system cannot afford such CPU usage amount.

You can always try to run the poisson surface reconstruction in the meshlab GUI, e.g. also see #17