search

gwenzhang / Voxel-Mamba

[NIPS'24] Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection
Apache License 2.0
85 stars 6 forks source link

The GPU memory requirement for model training #7

Open ouyangziyao opened 1 month ago

ouyangziyao commented 1 month ago

Hello, sir. This is an excellent work. I have much interest in the use of Mamba in point cloud detection, but I only have a single GPU with 12GB memory, much smaller than the memory you state in this paper. Can the model be trained on a single 12GB GPU?

gwenzhang commented 3 weeks ago

Sorry, we haven't specifically limited the GPU memory to 12 GB. However, you can try reducing the number of layers, halving the output resolution of BEV features, or increasing the size of input voxels.