search

tsingqguo / exposure-fusion-shadow-removal

We propose a new method for effective shadow removal by regarding it as an exposure fusion problem.
182 stars 29 forks source link

请问matlab代码计算rmse,指标达不到论文给的数 #45

Open yuanyufang opened 2 years ago

yuanyufang commented 2 years ago

作者您好： 我用您提供的去阴影结果，和matlab代码，计算rmse,发现无阴影区的数值，比论文中要大，比较迷惑，请问是什么原因？我是按照matlab代码中的要求，更改了无阴影，mask,和下载的去阴影结果数据路径，但是计算出来的数值较大，不知道是不是还要设置其他东西，我自己跑出来的结果去算rmse也是数值较大，很疑惑。

Dai-jl commented 2 years ago

我在另一篇论文的代码里看到原因，可以参考一下 “Note on the shadow removal evaluation code: We recently figured that the RMSE evaluation code that many papers have been using (including ours) is actually calculating Mean Absolute Error. We will retrospectively fix this in all our papers and suggest everyone to do the same.” https://github.com/cvlab-stonybrook/SID

yuanyufang commented 2 years ago

我在另一篇论文的代码里看到原因，可以参考一下 “Note on the shadow removal evaluation code: We recently figured that the RMSE evaluation code that many papers have been using (including ours) is actually calculating Mean Absolute Error. We will retrospectively fix this in all our papers and suggest everyone to do the same.” https://github.com/cvlab-stonybrook/SID

谢谢，我看了一下，这是另一个matlab代码，我用这个代码试一下，不过我在复现另一篇代码的时候，有用到你提到到的这个代码，最后计算的参数也比论文中要高，那篇论文是Mask-ShadowNet: Toward Shadow Removal via Masked Adaptive Instance Normalization

Dai-jl commented 2 years ago

我在另一篇论文的代码里看到原因，可以参考一下 “Note on the shadow removal evaluation code: We recently figured that the RMSE evaluation code that many papers have been using (including ours) is actually calculating Mean Absolute Error. We will retrospectively fix this in all our papers and suggest everyone to do the same.” https://github.com/cvlab-stonybrook/SID

谢谢，我看了一下，这是另一个matlab代码，我用这个代码试一下，不过我在复现另一篇代码的时候，有用到你提到到的这个代码，最后计算的参数也比论文中要高，那篇论文是Mask-ShadowNet: Toward Shadow Removal via Masked Adaptive Instance Normalization

一个计算的RMSE一个计算的是MAE，但在论文里都标的是RMSE，我用[https://github.com/cvlab-stonybrook/SID]里的代码跑这里给出的结果，跑出和论文差不多

yuanyufang commented 2 years ago

哦哦原来是这样子，之前都没发现给的是MAE，但是算的是MAE，在论文里面写RMSE，这能行？疑惑！行吧，反正作者也解释了，不懂MATLAB代码，看到就直接换了路径拿来直接用了。大佬，能加你联系方式吗，小菜硕想请教问题

yuanyufang commented 2 years ago

@.***，嘿嘿

---原始邮件--- 发件人: @.> 发送时间: 2022年6月14日(周二) 晚上10:15 收件人: @.>; 抄送: "yufang @.**@.>; 主题: Re: [tsingqguo/exposure-fusion-shadow-removal] 请问matlab代码计算rmse,指标达不到论文给的数 (Issue #45)

哦哦原来是这样子，之前都没发现给的是MAE，但是算的是MAE，在论文里面写RMSE，这能行？疑惑！行吧，反正作者也解释了，不懂MATLAB代码，看到就直接换了路径拿来直接用了。大佬，能加你联系方式吗，小菜硕想请教问题

我也是菜菜，就是先踩过这个坑而已，多交流少踩坑😂😂😂😂 你发个邮箱给我，我发你联系方式 @.***

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

yuanyufang commented 2 years ago

已发邮箱，哈哈