Serious work with the iso experiments

Started by 70MM13, June 28, 2018, 05:13:49 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

70MM13

New link for the mastered version:

http://drive.google.com/open?id=1RXZ5vrQggaX1EHi1Yeg0hA9gokJmnquc

I really pushed the limits of what I can do with extreme dynamic range using the iso experiments build on my 5D3.
Lots of content in shadow with a harshly lit subject.

No dual iso, no HDR, just ISO 200 dropped to 100.

The opening shot was done at 60fps using the standard crop rec 4k build.  I couldn't get usable results at high frame rates with the iso experiments build.  You'll notice extra noise in the opening shot.

Shot in 3k, downsampled to 1080 for smaller uploading.

Hevc RGB fullrange (0-255)

Download the file to see it properly.

I'm on both sides of the camera, so it's locked tripod shots when I'm in the frame...

I'm not sure if the https rule applies to goggle drive, but I removed the "s" from the link. If there's a problem, put it back in!

edit:  here's the december 2017 version of the iso research modules - use it with the Dec 2017 crop rec build:

http://drive.google.com/file/d/1MQQtNwqznPlURU5hs2GQuFqhUkzChIqN/view?usp=sharing

IDA_ML

Quote from: 70MM13 on June 28, 2018, 05:13:49 AM
I really pushed the limits of what I can do with extreme dynamic range using the iso experiments build on my 5D3.
Lots of content in shadow with a harshly lit subject.

No dual iso, no HDR, just ISO 200 dropped to 100.

Could you please explain in more detail how you get "extreme dynamic range" with that build, (link please)?  This would be very helpful to everyone like me willing to improve dynamic range of their cameras in the RAW video modes.  It does not have to be at high frame rates, 24 or 30 fps would be OK too.

Thank you in advance.

50mm1200s

Quote from: 70MM13 on June 28, 2018, 05:13:49 AM
http://drive.google.com/file/d/12-SEc4Xqd_n1ZzhtDXtyVzOHYt4YMABM/view?usp=sharing

File is way too big. Can you compress it? ProRes 422 or even 20MB/s h.264 would be nice.
Also, what are your reg values? Would be good to share on ADTG thread. With ISO changes and this could we finally beat Alexa? :p

50mm1200s

By the way, for anyone interested in more DR, check out Digicon Filter.

70MM13

It's already at 20 Mbps, but I will do a very short edit to make a smaller file.

The way I'm getting more dynamic range is by really lifting the shadows in post, thanks to the cleaner results I'm getting by using the ISO experiments build.  It makes a big difference for me.

I already posted my settings on the iso experiments thread, although they're nothing special.  I followed Alex's guidelines and I simply drop the adtg gain until the iso is half.  It works for me!

I use the December 2017 ISO experiments build.  It works well but as I noted above, it doesn't do the 50/60p modes as well as the default crop4k build.

I wish that this amplifier optimization would be taken seriously. For me, it's the biggest improvement of any of the ongoing projects.  There's no catch!  No aliasing, no complicated post processing, no surprises, just cleaner video!

Why isn't this being developed???

50mm1200s

Quote from: 70MM13 on June 28, 2018, 02:48:29 PM
I wish that this amplifier optimization would be taken seriously.

Indeed. I think this could be an option in the ISO menu. Maybe run a brutal force script to get the best DR and SNR regs for each camera model...
@dford compiled a version of 50D for me with the adtg branch and I can say it's stable enough to be on mainline.

70MM13

Here's a much smaller version of the clip.

It's a hack job, because I didn't want to reencode it, so I was forced to use keyframes from the original, which are few and far between...

Still, you will be able to see the nice clean shadowy details and very bright and nicely exposed subject matter!

http://drive.google.com/open?id=1ZR8ZxcVpJNy0Cw_zkGahxw6Ilz_mrZvC

allemyr

It looks good! I saw your example from a previous post and you can easily see a cleaner "grain" more noise free image when reducing the ISO this way. I lookad at the 100ish to 50ish ISO comparision.

Feels like a really good way! Was sceptic in the beginning but yes! It feels like Canon 5D3 or other cameras boost the ISO in a bad way so the noise gets a worse pattern so to say.

Great find! Would be really cool if there was a automatic setting for it, then I would use it!

70MM13

Thanks! I think so too.  The wide angle shot really shows what can be done.  That's serious dynamic range.

You're right, it's also a great way to get ISO 50 for extremely bright scenes.

I'm grateful to have even this cumbersome experimental version, but I have no doubt that there's far more potential improvement over the results I'm getting. There's no precision in how I do this. I just tinkered with it and settled on the "reduce by half" method while going through the ISO experiments thread.

It should be done properly!

If more of us are using it, development will surely follow.

IDA_ML

I have a question.  What is the difference between the ML Digital ISO, found on almost all cameras, where ISO can be reduced to -2 in 0.1 steps and the method described by 70MM13, when using the ISO Experimental build.  With ML Digital ISO on my 100D I can reduce the ISO to 80 (-0.3 setting) which is very close to the magical number of ISO 77 which should increase the dynamic range by 0.6 stops on this camera.  Is this the same as using the ISO Experimental build to achieve the same dynamic range improvement?

Thank you all who can provide a competent answer.

70MM13

The mastered version is now linked in the initial message.  If this doesn't convince people of the benefits of the ISO experiments, nothing will.