Can you provide a default config for 24G GPU device?

Pointcept / PointTransformerV3

[CVPR'24 Oral] Official repository of Point Transformer V3 (PTv3)

MIT License

583 stars 30 forks source link

Can you provide a default config for 24G GPU device? #40

Closed garrisonz closed 2 months ago

garrisonz commented 2 months ago

Thansk for your great work!

I have follow the README to install PTv3, and run the following command on GPU 3090, 24G

sh scripts/train.sh -p python -d scannet -c semseg-pt-v2m2-0-base -n semseg-pt-v2m2-0-base

torch.cuda.OutOfMemoryError error appears. Can you provide a default config requiring less gpu memory? Thanks Less GPU requirement quickstart example is more friendly to new birds. Thansk again

garrisonz commented 2 months ago

It work fine when I set batch_size to 2

Gofinge commented 2 months ago

Hi, setting batch_size to 2 will extremely harm the performance. Kindly reminding that you are running PTv2 instead of our efficient PTv3. BTW, typically, we need 4 GPUs to achieve the best scratch performance.