weird exposure/light switching while shooting raw on a mark 3

[manusch2]

first of all: thanks for ML, i've been using it for a while now, also on other cameras and it's just great!

let's get to the problem:

i don't really know where it comes from and unfortunately i couldn't find anything on this (i'm sorry if i missed something but i searched for quite a while... please tell me if it has already been discussed) but here's the problem: sometimes my 5dm3 switches the exposure of the picture. it only occurs when shooting raw, here's 2 examples - the quality is shit but you'll see it anyway: https://www.dropbox.com/s/p96ongfbw0cgw1q/5dmark3_problem.mp4

first i thought it would only occur when the general lighting situation in the frame changes but in the second example you can see that the problem even occurs when the lighting situation within the frame doesn't change.

i'd really be happy if anyone could help me out with that. thanks!

[hjfilmspeed]

Is auto iso on? or are you using ACR to process?
Auto ISO will not shift ISO smoothly with RAW since the smooth shift is from in cam digital gain which is not embedded in the raw file.

ACR has know issues with the whites, blacks, shadow, and highlight recovery sliders. In sequences they may shift a little. Lots of folks are either using the curves tab for adjustments or using the a log camera profile by vision color which will give you a log curve via camera calibration tab so it doesnt use the above offending sliders.

Also when shooting raw dont bother using isos in between 100 200 400 800 1600 3200 ect. ISO in between are in cam digital pulls and pushes. In good old fashion ML RAW video mode there is no meta data written to acknowledge the, what I call fake ISO, but new findings are leading to the fact that all ISOs on canon might be slightly manipulated by canon as well to reach exact stop increments.

It was once said to use th 160 320 640 ect because they are cleaner but this does NOT apply to ML RAW Dont bother using them unless you want a little more screen brightness. Other then that there useless unless your using h.264.

[manusch2]

thanks for this very quick answer!

the auto iso was never on so that's probably not the reason.

for ACR, yes i'm using it and i was also using those sliders (though i always tried not to exaggerate in using them, i know it can look ugly quite fast).
i suppose this is the canon a log profile by vision color that you were talking about http://vision-color.com/get-your-vision-picture-style/ ? looks interesting.
with the curves tool, you mean the curves tool within after effects? so i should just import the RAW sequence without any changes about highlights/shadows/blacks/whites and then go on with the curves tool to avoid those weird light shifts? can i still make changes on the other sliders?

for iso in steps 160, 320, 640 etc that's what i first learned to be best, well.. back then i was only shooting h.264 (but if i understand correctly, you're not saying this has anything to do with the exposure changing in my shots). thanks for that information, too.

[hjfilmspeed]

http://vision-color.com/visionlog/ this is the log profile its a camera profile for ACR or light room like Adobe Standard or camera standard in the camera profile tab. The link you posted was for a picture style that doesnt apply to RAW vid.

The curves tab is inside ACR. Thats the one i was talking about not the one in AE. Im saying if you want to make adjustments in ACR dont use the sliders I mentioned. use the sliders in the curves tab or adjust it manually. Supposedly you can mess with that and you wont get those little exposure shifts. If you mess with the highlight, shadows, whites and blacks sliders they will produce those little shifts that you saw. Im pretty sure that is what you were seeing. double check the forum but i think ever other slider inside ACR or LR is ok and wont produce that wierd shift. All this applies to LR too.

Yes and as for the iso settings, those shouldnt produce shifting your seeing but they wont accurately represent you true exposure in ML RAW video. if you use say iso 125, it is an in cam digital push of 100. If you brought up 2 raw clips in resolve, one shot at iso 100 and the other at iso 125 they should look identical in exposure because the in cam digital push was never recorded (if iso 125 is a digital push of 100 i kinda forgot) the same applies to 160 and iso 200 they should look the same in your raw editor. 160 is an in cam digital pull of 200. This is different then raw stills because the in cam pushes and pulls are recorded by the metadata i believe. ML RAW doesnt have metadata. h.264 rules dont apply any more with ML RAW. I only mentioned it because i saw 320 in one of your clips and i figured it was iso 320 which is a pull of 400. Try it. record 320 and 400 and bring it up in your raw editor should look the same if 320 is a pull of 400. not sure if this applys with the new raw format .mlv which does have metadata. i think its the same though.

[manusch2]

thanks a lot, i'll make tests with the vision log now!

also thanks for the iso issue, i knew that those in-between iso steps were pushes/pulls but i thought ML RAW would get that and it would still be the better decision to use those instead 100,200, 400 etc.

[a1ex]

This is different then raw stills because the in cam pushes and pulls are recorded by the metadata i believe.

In photo mode, the digital gain is *burned* in the CR2 raw data (not metadata). That's what we are going to fix with the new ISO tricks.

[manusch2]

Supposedly you can mess with that and you wont get those little exposure shifts.

seems like you're right! just tested it with visionLOG and the curves in ACR. guess that's what i'll work with from now on 

[hjfilmspeed]

In photo mode, the digital gain is *burned* in the CR2 raw data (not metadata). That's what we are going to fix with the new ISO tricks.

Wow the digital gain was burned in the raw stills too, thats crazzy! SOOO glad you guys are working on that! Let me know if I can help over here http://www.magiclantern.fm/forum/index.php?topic=10111.msg97459#new

^^^This is good reading too manusch2 ^^^