CMOS/ADTG/Digic register investigation on ISO

Started by a1ex, January 10, 2014, 12:11:01 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

ArcziPL

Quote from: timbytheriver on April 03, 2019, 12:35:58 PM
7) I shoot the clip at these settings. Now, I'm confused about the instructions to 'match the liveview exposure' What does this mean?
I'm not 100% sure but would understand it as follows: you are effectively reducing the sensitivity (ISO 400 -> xxx) so you should compensate it with with either increasing exposure time or aperture to match exposure between both photos, i.e. before and after adjustments.
M50.110 [main cam] | G7X III [pocket cam] | 70D.112 [gathers dust] | M.202 [gathers dust] | waiting for M5II

70MM13

matching exposure in the live view is critical.  keep in mind you are reducing the ceiling when reducing gain.  it'll show in the liveview scopes.  when you have it working properly, you will find that you can't go above whatever the new ceiling is, for example 95%.

if i had to guess what isn't happening for you is the photo capture/copy settings.  sometimes it just doesn't want to work, and i have to repeat the procedure a few times until it finally works.  ALL of the parameters will be filled in when it worked...

maybe i should make a little example video/tutorial for what i'm doing, if it might help somehow.

timbytheriver

Quote from: 70MM13 on April 03, 2019, 01:11:11 PM
maybe i should make a little example video/tutorial for what i'm doing, if it might help somehow.

@70MM13 That would be *very* much appreciated!  :D
5D3 1.1.3
5D2 2.1.2

70MM13

video uploaded, started a thread to discuss...

Danne

Hm, every time I see iso changing in these gain tests my own personal placebo journey into these regs comes to mind.
Not fully understanding how iso changing is relevant in this investigation I think the more correct way to work on these regs would be to don't let iso change at all. to check if there´s a way to actually get more dr from the sensor by pushing pregain and other related regs around and then check raw_diag.mo for proof. If iso ends up changing in these changes it's clearly something that affects the result much more than the actual pregain changes. So how much more I wonder?

70MM13

did you watch the video?

the results speak for themselves.

at the end of the day, i'm using it and getting indisputably good results, and it makes me happy.

just look at the last minute of the video if you refuse to watch the whole thing and ask yourself which result would you want on IDENTICAL images...

do results matter?  i'd call this the anti-placebo.

Danne

I say I can´t hardly tell what´s going on in certain parts of that vid. If I understand correctly you have the camera set to iso 200 and then reduce gain from 52 to 2. You then record a file but prior to this you change exposure manually viewing histogram? Next you turn these settings off and lower the iso 200 exposure matching the first example by viewing histogram?

So, key is that you change shutter in first example to get back lost light since it now reflects iso 111. This would mean you capture much more light into this particular set up mimicing exposure to the right only now it will look like it´s normally exposed. The second example you do the contrary. Apply the "normal" looking exposure but it will give an unfair result since shutter speeds are not the same anymore.

Question of a retest:
1 - Use the same shutter for both recordings. The first will be the normal looking histogram image but with reduced gain. The second recording will be with the same shutter but look brighter, more exposed to the right since you now turned of the reduced gain regs. I suspect in real life these two recordings will look very much the same when fixed in post.

2 - In post match the viewable brightness and then compare the grain and other seemingly better results.

70MM13







i followed your procedure.  200 was overexposed by a half stop.  i pulled it back in post.  it's better, but 111 still wins even when cheating in 200's favour.

and this indisputable improvement is free, and who knows how much better it can get with better adjustments with so many other registers?

we will never find out without exploring...

everyone can play!

timbytheriver

Reporting some operational issues arising from testing iso_regs on latest 4K crop-rec build (Experiments page).

Using 'Disable overrides' prompts message to restart (Help text) but upon restart values are not set back to default. Requires more restarts to happen as expected.

Using 'Copy Canon settings' nearly always fills only some values. Requires [often several] restarts to fill all values correctly.

5D3 1.1.3
5D2 2.1.2

lucoctober

I've tried it, and it works great. What I noticed, that the deep shadows now have green tint instead of magenta. Banding is still there, but overall nice improvement :)

timbytheriver

Quote from: 70MM13 on July 10, 2018, 07:43:49 PM

these images were created by loading the DNGs into mlv app, and dropping the dark strength to zero.


How did you load DNGs into MLVApp? I can't find anyway to do this!  ???
5D3 1.1.3
5D2 2.1.2

histor

By the way, iso research build doesn't work on 5d2 (2019Mar23.5D2212.zip). It can't power off the camera – both leds stay on for long (not 2-3 sec, as described – long enough to panic) [I needed some time to get the card door trick].

Another finding, which makes me exited. After reading the nice article from RawDigger developers (https://www.rawdigger.com/howtouse/iso-is-seldom-just-digital-gain) I ran into checking my own files from 5D2. With such a histogram it takes just several minutes to know, that the most interesting ISOs were 160, 320 and 640 – with clear signs of data division. All others look like pulled up (with periodic gaps in the histogram). That just proved what everybody told (but I'm a hard believer).

But that's not the end of the story. I continued checking files from my archive and found a batch of files with ISO100 and ISO50 (which is supposed to be complete fake) – with absolutely smooth histogram – no gaps, no spikes. How could it be so? It was hard to reproduce them: finally I grasped it. With EMF chip attached (telling f1.4) or an AF lens I get gaps in the histogram. Without electrical connection to the lens – histogram is nice and smooth. Peripheral lighting correction is disabled, so I can't guess what kind of correction is applied here. Should we avoid it?
It needs much more testing in the terms of noise and impact on real images. Taking a break now. Maybe, I'm completely wrong...


Continued:
I've checked results with Canon 50mm lens:
At f/1.4 – each 13th value in the histogram has a gap. It becomes less noticeable at closed diaphragms.
f/1.4 - each 13th
1.8-25
2.8-31
2.2-43
2.5-55
2.8-84(85)
3.2-70
I can't imagine any fair reason to use pull up at open apertures. Just maybe their lenses lacked light at open apertures and canon gays are simply cheating to satisfy specs? Looks rather probable.
In any case it would be nice to disable this pull up if somebody finds it.

70MM13

@timbytheriver
it must have been a typo...  i was clearly loading MLVs, not DNGs.

i think the reference might have been related to something i was outputting as DNG, therefore the easy mix-up.

sorry!

timbytheriver

@70MM13 No probs. I knew it was too good to be true!  :P
5D3 1.1.3
5D2 2.1.2

timbytheriver

Is there a knack/shortcut to turning the top wheel x hundred times to adjust each register in adtg_gui.mo?  ;)

E.g Change ADTG2 [8882]  to 0x13b (five minutes of turning!)



5D3 1.1.3
5D2 2.1.2

scrax

Quote from: timbytheriver on April 08, 2019, 01:07:48 PM
Is there a knack/shortcut to turning the top wheel x hundred times to adjust each register in adtg_gui.mo?  ;)

E.g Change ADTG2 [8882]  to 0x13b (five minutes of turning!)

can't you select each char in the number with arrows? On the module I use it works
I'm using ML2.3 for photography with:
EOS 600DML | EOS 400Dplus | EOS 5D MLbeta5- EF 100mm f/2.8 USM Macro  - EF-S 17-85mm f4-5.6 IS USM - EF 70-200mm f/4 L USM - 580EXII - OsX, PS, LR, RawTherapee, LightZone -no video experience-

timbytheriver

@scrax Thanks! That works a treat. Doh!  :P
5D3 1.1.3
5D2 2.1.2

timbytheriver

Is the CMOS[0] register in adtg_gui.mo the same as CMOS Gain in iso_regs.mo ?

I read of an adjustment to CMOS[0] to say 0x333, but will the CMOS Gain take the same input setting/value?
5D3 1.1.3
5D2 2.1.2

timbytheriver

@70MM13 Am I correct in thinking the reduced gain hack can be used in conjunction with dual_iso – swapping the recovery isos if that makes sense! Any benefits here?
5D3 1.1.3
5D2 2.1.2

timbytheriver

@Danne

When I use these settings in adtg_gui.mo from your earlier experiments I get keep getting a liveview/recorded image which looks like 2-strip technicolor – just greens and pinks.

analog gains changed to:
ADTG2 [8882] 0x13b
ADTG2 [8884] 0x13b
ADTG2 [8886] 0x13b
ADTG2 [8888] 0x13b

ISO 100 CMOS[0] 0x222

Elsewhere you mention that there are 8 registers to change on the 5D3...  :-\  Any ideas?

5D3 1.1.3
5D2 2.1.2

70MM13

that's exactly what i've been asking about for quite a while now...

it would be interesting to see what can be done.  if there's nothing preventing it from working, it may extend the range into the darkest shadows!

boldly go, brother...

timbytheriver

Have you tried it already? Or am I taking a leap of faith...?  :o
5D3 1.1.3
5D2 2.1.2

Danne

A1ex could advice here. My opinion atm is. Don't touch analog gains or cmos0 cause it will at best cause confusion. Instead enable raw_diag.mo and test pregain regs or other regs affecting highlights. Study raw_diag closely after changing and get used to the idea that only a slight refinement is likely to happen rather than "a new sensor" thing revolution.

timbytheriver

@Danne Do you mean for the dual_iso question – or the one previous (8 registers)?
5D3 1.1.3
5D2 2.1.2

timbytheriver

Have tried to get iso_regs and dual_iso working in tandem. Not working here. dual_iso menu shows only one iso... liveview shows no 'scanlines'.

Can they be made to work together for benefit? :)

5D3 1.1.3
5D2 2.1.2