search

LiJunnan1992 / DivideMix

Code for paper: DivideMix: Learning with Noisy Labels as Semi-supervised Learning
MIT License
529 stars 83 forks source link

Need For Kind Help in Paper's Replication #46

Open Christophe-Jia opened 2 years ago

Christophe-Jia commented 2 years ago

Hi li， I got a poor performance in both cifar-10/100 when the noise ratio is high than 80%, which mainly indicating 80% 90% in Table 1. All goes well with the hyperparameters in low noise condition. Could you kindly give me some suggestions for this problem I encountered ? More precisely, my results as follows: c10 0.8 best:75.7 last:73.5 c10 0.9 best:44.0 last:41.2 c100 0.8 best:52.7 last:50.9 c100 0.9 best:22.6 last:21.4

Christophe-Jia commented 2 years ago

With respect to the setting metioned in AppendixB Table 7,but still..../

jiawyGithub commented 1 year ago

The hyperparameters is important. my results as follows: c10 0.9 best:78.14 last:77.39 τ is set as 0.6. λu is set as 50.

Christophe-Jia commented 1 year ago

Thanks a lot for provideing hyperparameters setting! ANY recommended setting for 0.8 ?