search

snap-stanford / orca

[ICLR 2022] Open-World Semi-Supervised Learning
92 stars 12 forks source link

Hyperparameters for reproducing unseen accuracy on cifar100 #6

Closed ShichangZh closed 1 year ago

ShichangZh commented 2 years ago

Hello,

Thank you for the great work and for publishing the code! On CIFAR-100, I followed the instruction and ran the following line three times, without making any changes to the code. The maximum unseen accuracy I got was 0.4114, 0.4148, and 0.3890 for these three runs. May I ask how can I get the 43.0 reported in the paper? Any hyperparameters I need to change?

python orca_cifar.py --dataset cifar100 --labeled-num 50 --labeled-ratio 0.5
kleinzcy commented 2 years ago

Hi, @ShichangZh

Have you reimplement the results reported in the paper? We meet the same question as you. And I found that, setting mean_uncert as a fixed value (like 0.5) can get better results.

ShichangZh commented 2 years ago

No. I can't reproduce.