search

zju3dv / PGSR

code for "PGSR: Planar-based Gaussian Splatting for Efficient and High-Fidelity Surface Reconstruction"
https://zju3dv.github.io/pgsr/
Other
403 stars 22 forks source link

Memory consumption during evaluation #12

Closed Harryqu123 closed 1 month ago

Harryqu123 commented 1 month ago

Dear authors,

Thanks for your great work!

I am opening this issue since during my running of the code on my RTX 3090, I face an extremely high memory consumption during running the following line of code during evaluation(https://github.com/zju3dv/PGSR/blob/main/scripts/run_dtu.py#L24), and thus fail to perform the evaluation.

After my quick localizing of the place that leads to this high memory consumption, I find that it happens here:

rnn_idxs = nn_engine.radius_neighbors(data_pcd, radius=thresh, return_distance=False)

Could I please kindly know if you have any idea on the reason behond this?

Or possibly, could you please share the sklearn version that you use in your environment?

Looking forward to your reply and many thanks in advance.

danpeng2 commented 1 month ago

Hello, thank you for your interest in our work. This script runs purely on the CPU and does not use GPU computation. Do you mean that the system memory is insufficient? Our Sklearn version is 1.3.2.

Harryqu123 commented 1 month ago

Thanks for your timely response.

Yes I mean by system memory. In that case, could I please kindly know the system memory of your machine?

I am asking because, when I run the script, I find that it can occupy more than all the 126G + 117G system memory screenshotted below. I guess that it may be strange for the script to require such a large system memory. Sorry for bothering image

danpeng2 commented 1 month ago

00

We tested the scan24 sequence, and the maximum memory usage was 5GB.

Harryqu123 commented 1 month ago

Many thanks!

Yet the thing becomes a bit tricky here as we share the same python and sklearn version lol.

Anyway thanks for your timely information and have a good day ahead :)

I will try to figure some way to further debug my code.