search

JeongminB / E-D3DGS

[ECCV 2024] Official repository for "Per-Gaussian Embedding-Based Deformation for Deformable 3D Gaussian Splatting"
Other
109 stars 9 forks source link

Question on Memory #7

Closed 1305long closed 2 months ago

1305long commented 2 months ago

Hello, I ran out of 32G memory during the training. I want to know how much memory is needed for the complete training.

JeongminB commented 2 months ago

In my experience, training with a single 24G GPU has not been a problem. However, depending on the scene, an excessive number of Gaussians can lead to OOM issues.

The simple solution is to increase the values of densify_grad_threshold_fine_init and densify_grad_threshold_fine_after to reduce the total number of Gaussians. Alternatively, you might consider using a smaller value for lambda_dssim or disable it (set to 0).

Peamon commented 3 weeks ago

Hi, the processus takes a lot of RAM and i finally get a OOM (on pagefile.sys) cause no space left on my disk. Does the code have a memory leak or what parameter should i change to avoid this error ?

In your experience how much RAM memory use your process when training on cook_spinach ?

image