Open Zruto opened 3 years ago
Hi, Zruto.
During my review, I was also puzzled by this problem. At some period of time, I checked a blog that mentioned this idea. The idea that use an estimation curve to adjust image illumination and color may be inspired by this literature. I hope can help you.
@inproceedings{yuan2012automatic,
title={Automatic exposure correction of consumer photographs},
author={Yuan, Lu and Sun, Jian},
booktitle={European Conference on Computer Vision},
pages={771--785},
year={2012},
organization={Springer}
}
However, I do not remember where the blog is from. If someday I remember, I will pin the blog link here as a reference.
Best wishes.
-- Teng Li
Hi, Zruto.
During my review, I was also puzzled by this problem. At some period of time, I checked a blog that mentioned this idea. The idea that use an estimation curve to adjust image illumination and color may be inspired by this literature. I hope can help you.
@inproceedings{yuan2012automatic, title={Automatic exposure correction of consumer photographs}, author={Yuan, Lu and Sun, Jian}, booktitle={European Conference on Computer Vision}, pages={771--785}, year={2012}, organization={Springer} }
However, I do not remember where the blog is from. If someday I remember, I will pin the blog link here as a reference.
Best wishes.
-- Teng Li
Thank you so much. You've done me a big favor. Looking forward to this blog. I remember, the author mentioned this paper in related work,which is 【36】
Hi, @Zruto.
I found it. Here is the blog. I hope can help you.
https://blog.csdn.net/qq_36560894/article/details/108024748
Yep, literature [36] is referenced in the Zero-DCE paper.
Yuan and Sun [36] proposed an automatic exposure correction method, where the S-shaped curve for a given image is estimated by a global optimization algorithm and each segmented region is pushed to its optimal zone by curve mapping. Different from [36], our Zero-DCE is a purely datadriven method and takes multiple light enhancement factors into consideration in the design of the non-reference loss functions, and thus enjoys better robustness, wider image dynamic range adjustment, and lower computational burden.
Best wishes.
-- Teng Li
公式可以推出来,映射曲线为二次曲线f(x)=ax^2+bx+c,满足条件f(0)=0和f(1)=1以及(0,1)区间内单调增,就可以得到c=0,a+b=1,a属于区间[-1,1]
@Zruto @Ethanwolfie @litun5315 想问下各位我测试过曝光图片发现性能不是很好,但是我看到训练集里是多曝光图片,感觉应该同时处理低曝光和过曝光,各位有试过过曝光测试吗?还是说需要修改enhance模块?谢谢
您好,想问您关于光照增强曲线是受什么启发设计出来的,仅是编辑软件的曲线调整吗?有没有具体的参考文献呢?