Low Acc when training ssm_2d

ethanbar11 / ssm_2d

More dimensions = More fun

21 stars 2 forks source link

Low Acc when training ssm_2d #4

Open anhtienng opened 3 months ago

anhtienng commented 3 months ago

Hi,

I am using your existing code and setting to run ssm_2d on CIFAR-10, but the top-1 accuracy is low.

best top-1: 57.20, final top-1: 56.30

I'm running python main.py --model vit --dataset CIFAR10 --emb ssm_2d. Are there any configs that I should change ?

Thank you.

ethanbar11 commented 3 months ago

We didn't test it on CIFAR-10 (the results aren't shown in the paper) and I don't remember the hyper-parameters you need to use. I'd suggest starting from one of the other ones, CIFAR-100, Tiny-ImageNet or ImageNet