search

VITA-Group / EnlightenGAN

[IEEE TIP] "EnlightenGAN: Deep Light Enhancement without Paired Supervision" by Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, Zhangyang Wang
Other
918 stars 200 forks source link

the psnr package #2

Closed YangQun1 closed 5 years ago

YangQun1 commented 5 years ago

May I ask, where can I get the psnr package?

YangQun1 commented 5 years ago

And, could you please upload the pre-trained model?

yifanjiang19 commented 5 years ago

psnr package is not used so I delete it and we found that PSNR/SSIM is not suitable for low light enhancement task. I'll upload pre-trained model later.

YangQun1 commented 5 years ago

@yueruchen thanks，I also found that PSNR or SSIM is not suitable. but, when I use NIQE to evaluate the results, I found that the NIQE score is not stable. May I ask, what model parameters are you use? I use the NIQE code and parameters form this GitHub repo https://github.com/dsoellinger/blind_image_quality_toolbox.git

YangQun1 commented 5 years ago

for example, the CLAHE's results are more poorly in visual quality, but its NIQE score is lower.

laomaup commented 5 years ago

thank you,but I can't find the file requirements.txt.

yifanjiang19 commented 5 years ago

@YangQun1 You are right, there isn't suitable metric in this area, NIQE is just a comparably good one. So we use visual results, user study, and non-reference metric to evaluate our method.

yifanjiang19 commented 5 years ago

@laomaup Sorry, I've updated it already.

yifanjiang19 commented 5 years ago

@YangQun1 NIQE code can be found in its authors' website.