FRSP makes heavily overexposed areas black

Started by elenhil, August 08, 2022, 07:15:00 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

elenhil

I've been trying to shoot a timelapse with FRSP, and some of the frames got the sun in them. Since I was using AETTR, the sun was, naturally, overexposed. But compared to regular photos or videos, it was black!

Is this a side effect of FRSP? Or of converting the resulting MLV container into separate DNGs? Can it be fixed in post-processing?

tupp

The "black sun" effect is a well known phenomenon with CMOS camera sensors -- it has nothing to do with Magic Lantern.

Here is one way to fix it in post.

elenhil

Quote from: tupp on August 08, 2022, 08:10:17 AM
The "black sun" effect is a well known phenomenon with CMOS camera sensors -- it has nothing to do with Magic Lantern.

Here is one way to fix it in post.
Strange. I've been shooting stills with other Canon cameras for quite some time, and never encountered this phenomenon; the overexposed areas were always white.

Anyway, might there be some other means of fixing it without tracking and masking involved? Something at the DNG conversion stage, perhaps?

Walter Schulz

This error "correction" is a main task in sensor data manipulation. It won't be seen in most cases (there are some well known exceptions which got fixed in a hurry) but ML's way to treat data is a bit different. You may say "not as sophisticated".

See https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6386896/
QuoteThe black sun effect is an unwanted effect that most manufacturers remove through a post-image process by replacing the affected pixel value with the surrounding pixel value. Thus, it was very

elenhil

Quote from: Walter Schulz on August 08, 2022, 08:43:21 AM
This error "correction" is a main task in sensor data manipulation. It won't be seen in most cases (there are some well known exceptions which got fixed in a hurry) but ML's way to treat data is a bit different. You may say "not as sophisticated".

See https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6386896/
Sorry, your reply has left me somewhat mystified. Could you please expand on it? Do you mean to say that a FRSP (а very much "ML" way of acquiring data from the sensor) will always be prone to this phenomenon, while a "regular" photo, or even video (a "Vanilla" way of doing it) won't?

Walter Schulz

Exactly.
Don't know if there may be a way to apply said data manipulation into FRSP. This is dev area.


Skinny

I always get pink sun in my footage, and it is easy to mask out... create a layer filled with color (just use color picker and sample the brightest area near the sun), then add mask (slightly bigger) to this new layer and feather it a little. Done.
I don't even track the mask sometimes when there is not a lot of motion. But you can track it with auto tracking, it won't take much time. It is just one object with fixed size and precise position doesn't really matter, so you can loosely track it and it will be ok.


It would be cool if some day MLV App will be upgraded with black sun detection/replacing with surrounding color..