RankIQA.PyTorch icon indicating copy to clipboard operation
RankIQA.PyTorch copied to clipboard

实现的疑问

Open JDanielWu opened this issue 4 years ago • 2 comments

你好 我想问下,按照原论文来说,应该有个FT(fine tune)过程,就是失真图片进行rank 孪生网络训练后,要用实际的有评分的图片进行fine tune,这样最后输出的是一个评分。你的实现我看到有3个任务,排序对比损失任务,分布的推土距离损失任务,回归损失任务。排序对比损失任务 我感觉比较接近第一步部分,FT 部分是回归损失任务? 我看样例里AVA好像也没用到回归损失。希望有空能解答下,谢谢。

JDanielWu avatar Jun 02 '21 09:06 JDanielWu

the same issue, want to implement original paper in Pytorch but having trouble in understanding in detail like does the siamese network work in the same way, how it is trained and where is the Ranking dataset. Would be grateful if anyone replies. Thank you!

ayushrox avatar Jun 06 '21 06:06 ayushrox

对,ava就是rank的时候用用,rankiqa本质是给tid和live设计的,所以前边rank之后,再在tid上finetune regression任务

YilanWang avatar Apr 21 '22 02:04 YilanWang