hongxin001 / JoCoR

CVPR'20: Combating Noisy Labels by Agreement: A Joint Training Method with Co-Regularization

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Can you provide the best co-lambda for cifar-10 and cifai-100, please

tzhxs opened this issue · comments

commented

Actually, it depends on the datasets, noisy rate and basic models. Generally, the best lambda is between 0.5-0.9;

For example, in my experiment, the best lambda for CIFAR10 with 50% noise is 0.9. When it comes to 80% noise, the best is 0.65.

If you change the basic model to ResNet or others, you need to try different lambda for the best.