[FFmpeg-devel] [PATCH] libavfilter: Add derain filter init version--GSoC Qualification Task.

[xwmeng at pku.edu.cn]

> -----原始邮件-----
> 发件人: "Pedro Arthur" <bygrandao at gmail.com>
> 发送时间: 2019-04-11 00:12:09 (星期四)
> 收件人: "FFmpeg development discussions and patches" <ffmpeg-devel at ffmpeg.org>
> 抄送: 
> 主题: Re: [FFmpeg-devel] [PATCH] libavfilter: Add derain filter init version--GSoC Qualification Task.
> 
> Hi,
> Em ter, 9 de abr de 2019 às 22:42, <xwmeng at pku.edu.cn> escreveu:
> >
> > Yes, I use the espcn model for deraining as the initial version as it's a easier way to implement the filter, although the paper proposes it for super-resolution. And the model does have some effect on deraining project. While, it is just the first version. I will use more suitable and more powerful model for derain filter according to the latest models proposed in derain task, and I will upload the new model soon.
> >
> 
> There is no problem in using the espcn infrastructure to start
> learning, but I still can't see how using the espcn model fits your
> purpose.
> The espcn model should output an image bigger than the input, why
> would a derain filter do that?
> 
> Also you did not provide any data regarding the results you might have
> obtained neither the model file and appropriate input/reference so we
> can test it.
> What are the results you get so far (psnr before/after applying the filter)?
> _______________________________________________

We made some modifications on the original ESPCN model, such as change the input image from one channel(Y) to three channels(RGB) and remove the up-sampling procedure. The model file has been uploaded in https://github.com/XueweiMeng/derain_filter and you can download the training/testing dataset from http://www.icst.pku.edu.cn/struct/Projects/joint_rain_removal.html. I didn't save the PSNR/SSIM score during the training and evaluating process. So the data will be uploaded later.

Xuewei

_______________________________________________

> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".