danfenghong
commented
2 years ago Thank you very much for your interest to our work! The weight_decay should 5e-3 rather than 0. You can refer to the parameter selection we provided. Cheers, Danfeng
loveswine @.***> 于2022年2月6日周日 15:55写道：
Dear Dr.Hong: Thanks a lot for you sharing code. I ran your demo for training on Indian dataset twice, the test accuracy is about 76%, which is not as accuracte as your paper, here are the results and arguments, [image: 屏幕截图 2022-02-06 151032] https://user-images.githubusercontent.com/53184094/152671083-2aaaa2b7-2579-4aaf-8c52-3a0c4b27cf3c.png
And i also run your spectralformer model in my training code framework for convenient comparing, but the accuracy is relatively low. So i look for the difference in codes and have a few questions：①Does the spectralformer sensitive to whitening algorithm? ②I found the training process is stable and convergent in 100 epochs, why set 300 or 600 epochs ? it may be time consuming.
— Reply to this email directly, view it on GitHub https://github.com/danfenghong/IEEE_TGRS_SpectralFormer/issues/10, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFL2GZQQXK3ONKZPVAZNHZDUZYSQZANCNFSM5NU775VA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
You are receiving this because you are subscribed to this thread.Message ID: @.***>
Dear Dr.Hong: Thanks a lot for you sharing code. I ran your demo for training on Indian dataset twice, the test accuracy is about 76%, which is not as accuracte as your paper, here are the results and arguments, is there anything wrong i set?
And i also run your spectralformer model in my training code framework for convenient comparing, but the accuracy is relatively low. So i look for the difference in codes and have a few questions：①Does the spectralformer sensitive to whitening algorithm? ②I found the training process is stable and convergent in 100 epochs, why set 300 or 600 epochs ? it may be time consuming.