accuracy of Dense Net is lower than accuracy in your paper

JianshuZhang / WAP

Watch, Attend and Parse for Handwritten Mathematical Expression Recognition

250 stars 82 forks source link

accuracy of Dense Net is lower than accuracy in your paper #11

Closed ducanh841988 closed 5 years ago

ducanh841988 commented 6 years ago

Hi Jianshu! I retrain Dense Net Decoder. I got 17% of WER and 44% of Exp Rate on validation and 17.05 WER and 42.6% of Exp Rate on Testing set (CROHME 2014 test set). It is lower than accuracy in your paper. Is this version same with the version of your paper? I want to get the same accuracy as in your paper, could you give me advises? Thank a lot. Anh

JianshuZhang commented 6 years ago

Hi, good to see you! The result presented in my paper is an ensemble of 5 models. You can try that, I believe your result is just a single model.

ducanh841988 commented 6 years ago

Thank you for your quick answer. Is it consistent with a single model in your experiment? I will try your model with my data generation. Thank you.

JianshuZhang commented 6 years ago

Yes. This is my source code.

发自我的 iPhone

在 2018年8月20日，10:11，LE DUC ANH notifications@github.com 写道：

Thank you for your quick answer. Is it consistent with a single model in your experiment? I will try your model with my data generation. Thank you.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.