when training Transfusion_LC, CUDA out of memory. （NVIDIA 2080 8G）

XuyangBai / TransFusion

[PyTorch] Official implementation of CVPR2022 paper "TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformers". https://arxiv.org/abs/2203.11496

Apache License 2.0

619 stars 76 forks source link

when training Transfusion_LC, CUDA out of memory. （NVIDIA 2080 8G） #34

Open wangyd-0312 opened 2 years ago

wangyd-0312 commented 2 years ago

hi, Mr. Bai I have trained Transfusion_L, and got the fusion_model.pth. When I want to train Transfusion_LC, I get this error "CUDA out of memory", I use NVIDIA 2080 8G, and I change sample_per_gpu=1、lr=0.0025, and the database is nuscenes mini. I want to know if 2080 8G can run the Transfusion_LC?

Looking forward your replay, thank you. 捕获1

XuyangBai commented 2 years ago

Hi, I am afraid 8G is not enough for training TransFusionLC, you should find GPUs with larger memory (i.e. 16G should be enough for LC).

wangyd-0312 commented 2 years ago

Thank you for your replay!!

wqueree commented 1 year ago

@XuyangBai is a single GPU sufficient for training L and LC? Or is distributed training across many 16G GPUs required?