Question about the ensemble method in the code

Question

Question about the ensemble method in the code

Closed this issue 6 years ago · 6 comments

Hi Jianshu! I have read your paper(it's a great job), and I have also run your code in my GPU(GTX 1080Ti). But I only achieved 42% recognition rate using the densenet. Your paper says that you use 5 models to get a better result(50%), I'm wondering if your released code has implemented the ensemble method?

SuperHan777 commented 6 years ago

Thanks!

Answer 1 · 2018-05-07T02:56:28.000Z

The ensemble codes will be released in soon future, waiting for the notification of another paper's submission

Answer 2 · 2018-05-07T06:26:09.000Z

Is it normal for me to use one model to get 42% recognition rate? Is the ensemble method makes such a great improvement?

Answer 3 · 2018-05-07T06:51:59.000Z

single model achieves about 44-45% ExpRate. You can try training 5 models to see the average performance. xysszjs@mail.ustc.edu.cn From: SuperHan777 Date: 2018-05-07 14:26 To: JianshuZhang/WAP CC: Jianshu Zhang; Comment Subject: Re: [JianshuZhang/WAP] Question about the ensemble method in the code (#7) Is it normal for me to use one model to get 42% recognition rate? Is the ensemble method makes such a great improvement? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 4 · 2018-05-11T09:02:16.000Z

Hi Jianshu!
I have read your DenseNet code and paper recently. It seems that the code doesn't contain the multi-scale attention branch. I'm not very sure about this, could you tell me? thanks!

Answer 5 · 2018-05-11T09:22:00.000Z

Yes, only single-scale attention branch. The paper was recently accpeted, we didn't release the code yet. We will release the multi-scale code in soon future. xysszjs@mail.ustc.edu.cn From: SuperHan777 Date: 2018-05-11 17:02 To: JianshuZhang/WAP CC: Jianshu Zhang; Comment Subject: Re: [JianshuZhang/WAP] Question about the ensemble method in the code (#7) Hi Jianshu! I have read your DenseNet code and paper recently. It seems that the code doesn't contain the multi-scale attention branch. I'm not very sure about this, could you tell me? thanks! — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.