alpha_mix_active_learning icon indicating copy to clipboard operation
alpha_mix_active_learning copied to clipboard

Published 20 hours ago • verified publisher icon AminParvaneh

Accuracy problem on CIFAR100 dataset

Open NikosKats opened this issue 1 year ago • 0 comments

I got the mentioned results for the dataset MNIST, but when I use the CIFAR100 dataset with these parameters:

--data_name CIFAR100 --data_dir your_data_directory --log_dir your_log_directory
--n_init_lb 1000 --n_query 1000 --n_round 10 --learning_rate 0.001 --n_epoch 2000 --seeds 1 --model vit_small --pretrained_model true --fine_tune_layers 1
--strategy AlphaMixSampling --alpha_opt

I get maximum accuracy of 0.3843 instead of 52.5 as mentioned in the paper. Can you please clarify if I use the correct parameters or not?

NikosKats avatar Apr 08 '23 08:04 NikosKats