search

PRBonn / lidar-bonnetal

Semantic and Instance Segmentation of LiDAR point clouds for autonomous driving
http://semantic-kitti.org
MIT License
915 stars 204 forks source link

training/inference on laptop #79

Closed amkurup closed 3 years ago

amkurup commented 3 years ago

Hi, Is it possible to train or run inference on a single laptop (System76 - GeForce GTX 1650 - 4Gb)? Both these now return "Cuda out of memory error". nvidia-smi shows that most of the memory is free.

Is there a workaround for this or does it have to be done on a stronger system?

kosmastsk commented 3 years ago

Try to reduce the batch size as more as possible and see if you can run the training process. But in my opinion, this GPU is not enough due to the large size of pointcloud data.

amkurup commented 3 years ago

I have tried reducing the batch size and also tried to use only sequence 01 for training (as a test). Still the same issue.

kosmastsk commented 3 years ago

Are you using darknet53 as the backbone of your model? This one requires more memory, than squeezeseg e.g., so you could try a different backbone

amkurup commented 3 years ago

I am using darknet53. Thanks, will try out squeezeseg!

amkurup commented 3 years ago

squeezesegV2 works on this system. This will do for now! Thanks