when compare the middle exposure frame with aligned gt，I find that there is a significant misalignment that the highlight of middle exposure shot is (0.9725, 0.9725, 0.9725) while gt is (1, 1, 1). Then I check short/middle/long shots and find that the max value of all frames reach (0.9725, 0.9725, 0.9725) instead of (1, 1, 1). Can anyone explain or fix this problem?
This problem definitely affects the estimation of true gamma.
Hello qiushizai,
That maximum value your are seeing in your input images is caused by the presence of a dark current I_0 during the image capture (that is a constant offset representing the black point), which in turn causes sensor saturation to occur when (I_max − I_0)*g electrons are collected by the sensor (that is basically the "full well" electron capacity limit) for a certain sensor gain g.
One could normalize that [0, I_max-I_0] range to be in the [0,1] range. However that is a design choice, and we decided to present the data with no further data normalizations, respecting the original quantized levels as much as possible, and let the participants come up with design choices that result in optimal performance on that regard as well.
So in summary: those values are not "misaligned", they are just saturated. If you are referring to another type of misalignment, please point at some specific examples so that I can take a closer look at them and see if there is any other issue going on.
Best regards,
Edu.
So in summary, the loss of white point is just because you remove the digital gain(or normalize) of black level correction(BLC) in ISP pipline. Thanks for your patient answer~
Posted by: qiushizai @ Feb. 10, 2021, 2:02 a.m.