MCG-NJU / SparseOcc

[ECCV 2024] Fully Sparse 3D Occupancy Prediction & RayIoU Evaluation Metric
https://arxiv.org/abs/2312.17118
Apache License 2.0
235 stars 16 forks source link

How much vram do I need to train this model? #13

Closed leavesullic closed 4 months ago

leavesullic commented 5 months ago

thanks for the great work~ The README model zoo tells that the training cost is 1d4h, ~12GB Memory. But it runs out of memory when I train the r50_nuimg_704x256_8f model on one RTX4090. I did not change anything in r50_nuimg_704x256_8f.py

What parameters should I decrease to train this model on a 24g vram machine?

afterthat97 commented 5 months ago

you need 8 GPUs

leavesullic commented 5 months ago

Hmmmm, are 4 gpus enough to train?

afterthat97 commented 5 months ago

maybe

afterthat97 commented 5 months ago

if that still not works, you can reduce num_frames to 6