Open mengmengda1127 opened 3 years ago
Any updates on this? Would be great to hear.
Too busy to find the source testing code. I will check it out soon.
发自我的iPhone
在 2020年12月25日,下午3:51,Siddharth Shrivastava notifications@github.com 写道:
Any updates on this? Would be great to hear.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.
Alright, Thanks!
The psnr value I tested is the highest and 0.03 lower than the paper
@mengmengda1127 I collect 791 paired images from Part2 of SCIE dataset, but 767 paired low/normal light images are reported in the paper as test dataset. I cannot find which images are discarded. Would you please to share the index or filenames of the discarded images? Thank you very much.
@mengmengda1127 I collect 791 paired images from Part2 of SCIE dataset, but 767 paired low/normal light images are reported in the paper as test dataset. I cannot find which images are discarded. Would you please to share the index or filenames of the discarded images? Thank you very much.
I collect 786 paired image from Part2 of SCIE dataset@PeeBar @mengmengda1127. And I test that 786 images which are resized into 1200x900 uniformly using the snapshot of Epoch99.pth. I get a psnr of 16.1738. The following is my psnr code
def psnr(img1, img2, PIXEL_MAX=255.0):
img1 = np.float32(copy.deepcopy(img1))
img2 = np.float32(copy.deepcopy(img2))
mse = np.mean((img1 - img2)**2)
if mse == 0:
psnr = 100
else:
psnr = 20 * math.log10(PIXEL_MAX / math.sqrt(mse))
return psnr
The img1 and img2 are both resized into 1200x900 before being put into function psnr.
Hi author,
Currently I’m reproducing the results in Table 2 (which are quantitative comparisons) of your paper. However, although I prepared the 767 paired low/normal light images of SICE dataset Part2 subset following the instruction in 4.2, my reproduced PSNR/SSIM values no matter from ZeroDCE nor from other compared methods are always lower than those in Table 2 (shown in the following table). I doubt the reason is either we create different SICE datasets or use different evaluation ways (e.g. different image resizing methods). So can you kindly provide your prepared SICE dataset and the evaluation code as well? Thank you.
Table 1 The interpolation method is cv2.INTER_CUBIC.
ZERO-DCE EnlightenGAN LIME PSNR 16.1580 16.0603 15.8231 SSIM 0.4866 0.5374 0.4813
hi, could you please share your code which is used to calculate psnr and ssim? @Li-Chongyi @PeeBar @Zruto
Hi author,
Currently I’m reproducing the results in Table 2 (which are quantitative comparisons) of your paper. However, although I prepared the 767 paired low/normal light images of SICE dataset Part2 subset following the instruction in 4.2, my reproduced PSNR/SSIM values no matter from ZeroDCE nor from other compared methods are always lower than those in Table 2 (shown in the following table). I doubt the reason is either we create different SICE datasets or use different evaluation ways (e.g. different image resizing methods). So can you kindly provide your prepared SICE dataset and the evaluation code as well? Thank you.
Table 1 The interpolation method is cv2.INTER_CUBIC.
ZERO-DCE EnlightenGAN LIME PSNR 16.1580 16.0603 15.8231 SSIM 0.4866 0.5374 0.4813
Hello, when I evaluate the code, there is no file containing the dataset DICM、LIME、MEF infered in paper "Zero-Reference Deep Curve Estimation (ZeroDCE) for Low-Light Image Enhancement", can I ask for a help?
Hi author,
Currently I’m reproducing the results in Table 2 (which are quantitative comparisons) of your paper. However, although I prepared the 767 paired low/normal light images of SICE dataset Part2 subset following the instruction in 4.2, my reproduced PSNR/SSIM values no matter from ZeroDCE nor from other compared methods are always lower than those in Table 2 (shown in the following table). I doubt the reason is either we create different SICE datasets or use different evaluation ways (e.g. different image resizing methods). So can you kindly provide your prepared SICE dataset and the evaluation code as well? Thank you.