mit-han-lab / bevfusion

[ICRA'23] BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation
https://bevfusion.mit.edu
Apache License 2.0
2.29k stars 413 forks source link

Some problems about test #365

Closed usugami closed 1 year ago

usugami commented 1 year ago

I know that BEV Fusion requires multiple GPUs such as A6000 to train the model in order to ensure sufficient computing power and memory. However, I want to know if the trained model can be deployed on a small computer such as Xavier NX. If not, what is the minimum required configuration for it to run? Thanks a lot.

YaqinLong commented 1 year ago

I tried to train the bevfusion on Xavier NX, but it failed because out of memory.

VeeranjaneyuluToka commented 1 year ago

I was able to run the inference of camera or LiDAR alone models using 8GB memory GPU card but Camera+LiDAR failed to run in 8GB and ran successfully using 24GB GPU card. I think you can run camera+LiDAR model using 12 or 16GB memory GPU card also, however i have not tried.