Wentao Zhu

> This is will be a new feature request, as per current design, it's going to be a separate `AudioReader` and a wrapper reader for both video/audio, e.g. `MediaReader`. Are...

Or maybe the conversion between reg_resample and itk ResampleImageFilter?

Thank you so much, Mostafa! @MostafaDehghani and @anuragarnab

还是达不到0.78，只能0.775+

tensorflow for deep learning models, and matlab for data processing

> 1.五折交叉验证+多模型融合 2.如果GPU资源够多可以使用大模型 3.可以用同义词替换、随机插值、回译方法数据增强。 > 我用了百度api进行回译，将回译数据和原数据合并去重后，训练，测试效果确实有提高。我会上传一下合并后的数据集 使用回译数据训练，模型的超参数怎么设置呢？epoch number？early stop的数目？等其他参数？谢谢！

There is a pdf in the folder.

I added JsonUtils.py in the folder. You can download it and create utils folder, then put it into the folder.

https://github.com/wentaozhu/recurrent-attention-for-QA-SQUAD-based-on-keras/blob/master/JsonUtil.py