OOM when evaluate_dtu_mesh

[fangchuan]

Hi, thanks for your wonderful work! I'm trying to repeat the quantitative results in your paper on a single A100 GPU. I am able to generate the tsdf fusion mesh in voxel_size=0.002, but I fail to run evaluate_dtu_mesh.py and was killed in the cull_mesh() function: https://github.com/autonomousvision/gaussian-opacity-fields/blob/c3b4c9ee75866bedcca1d7b2b4dac31aee20be59/evaluate_dtu_mesh.py#L134

It takes over 1000 RAM and then is killed. Could you give me some suggestion or solution for this problem? @niujinshuchong Thanks in advance!

[ruili3]

Hi, thanks for your wonderful work! I'm trying to repeat the quantitative results in your paper on a single A100 GPU. I am able to generate the tsdf fusion mesh in voxel_size=0.002, but I fail to run evaluate_dtu_mesh.py and was killed in the cull_mesh() function:

https://github.com/autonomousvision/gaussian-opacity-fields/blob/c3b4c9ee75866bedcca1d7b2b4dac31aee20be59/evaluate_dtu_mesh.py#L134

It takes over 1000 RAM and then is killed. Could you give me some suggestion or solution for this problem? @niujinshuchong Thanks in advance!

Hi,

How do you manage loading dtu_cameras and gt_points here https://github.com/autonomousvision/gaussian-opacity-fields/blob/c3b4c9ee75866bedcca1d7b2b4dac31aee20be59/evaluate_dtu_mesh.py#L148 I can't find the corresponding files required in load_dtu_camera. How do you solve this? I can not run to your position otherwise we can take a look at it together.

[fangchuan]

Hi, thanks for your wonderful work! I'm trying to repeat the quantitative results in your paper on a single A100 GPU. I am able to generate the tsdf fusion mesh in voxel_size=0.002, but I fail to run evaluate_dtu_mesh.py and was killed in the cull_mesh() function: https://github.com/autonomousvision/gaussian-opacity-fields/blob/c3b4c9ee75866bedcca1d7b2b4dac31aee20be59/evaluate_dtu_mesh.py#L134

It takes over 1000 RAM and then is killed. Could you give me some suggestion or solution for this problem? @niujinshuchong Thanks in advance!

Hi,

How do you manage loading dtu_cameras and gt_points here

https://github.com/autonomousvision/gaussian-opacity-fields/blob/c3b4c9ee75866bedcca1d7b2b4dac31aee20be59/evaluate_dtu_mesh.py#L148

I can't find the corresponding files required in load_dtu_camera. How do you solve this? I can not run to your position otherwise we can take a look at it together.

You need to download the necessary data on DTU

[niujinshuchong]

Hi, what do you mean by 1000 RAM? do you mean 1000GB?

[Shhfdsdf]

it seems like i get the same error,i use the uncull mesh to eval before,it's still be killed

[Shhfdsdf]

it seems like i get the same error,i use the uncull mesh to eval before,it's still be killed

after i removed these few lines of code,it works!but I'm not sure if the result is accurate

python dtu_eval/eval.py --data exp_dtu/release/scan24/test/ours_30000/tsdf/tsdf_aligned.ply --scan 24 --mode mesh --dataset_dir dtu_eval/Offical_DTU_Dataset --vis_out_dir exp_dtu/release/scan24/test/ours_30000/tsdf done: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 9/9 [01:51<00:00, 12.34s/it] 0.4626517840226122 0.47273726170521585 0.467694522863914