Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

New RED Color Science - Test Pilot Feedback Thread

Niko Pueringer said:
I'd like to pose this question again, as I don't feel like it really got answered; Based on what I'm seeing, we're seeing less noise because the blue channel is turned down in strength to make for accurate white balance temperatures. (If this is not the case, please correct me).

If the blue channel is turned up to the same strength to match what an earlier build would look like, the noise is about the same then?

This question is based off the images Nate posted a few pages back...
Click to expand...

Not really...

Graeme has re-written the entire color science. Everything is different. The color temp now matches a color meter and WB on camera, the monitor output now matches both. There is measurably less noise. To try and simplify what Graeme has done to one convenient sound bite does not do justice to the complexity of the new system he has put in place. You can guess all you want. Just be prepared not to hear what he did and why it works.

Jim
 
Jannard said:
Not really...

Graeme has re-written the entire color science. Everything is different. The color temp now matches a color meter and WB on camera, the monitor output now matches both. There is measurably less noise. To try and simplify what Graeme has done to one convenient sound bite does not do justice to the complexity of the new system he has put in place. You can guess all you want. Just be prepared not to hear what he did and why it works.

Jim
Click to expand...

Snap.
 
As far as I can see, if you take a while balance picker on a properly neutral grey / white object, neither the old nor new color science is any warmer or cooler than it should be, and that white picked area is white.

There's no way I could explain what I did anyway - it's taken me over a year of learning, experimenting and programming to figure all this out, and that's after about 6 months learning and work on the original. As someone who is generally good at math and science stuff, I've got say that color science can be quite hard, and one of the trickiest things I've tried to do. (perhaps other than making 4k24fps@10mb/s work).

Graeme
 
Graeme Nattress said:
As far as I can see, if you take a while balance picker on a properly neutral grey / white object, neither the old nor new color science is any warmer or cooler than it should be, and that white picked area is white.

There's no way I could explain what I did anyway - it's taken me over a year of learning, experimenting and programming to figure all this out, and that's after about 6 months learning and work on the original. As someone who is generally good at math and science stuff, I've got say that color science can be quite hard, and one of the trickiest things I've tried to do. (perhaps other than making 4k24fps@10mb/s work).

Graeme
Click to expand...

Hahahaha.... :rofl:

I still haven't heard the answer about the whiskey...

:devil: Peter
 
I like the analogy to film stocks - we don't know nor care what Kodak changes to achieve a new stock, we care how things look when the film is processed - and while it would be good to have documentation on how this new color science responds in various conditions one would think that this is something that will come from lots of testing, which is what the Alpha and Beta build is about. Once there is enough feedback we will hopefully have a good view of what these changes mean in practice, it may be a bit easier to explain to wary DOP's in the meantime that they can try the new stuff and test their own results, or just shoot as you have been and keep showing them the tests that come up until the decide to dip their feet in (from my experience they tend to be a reasonably curious bunch) - at least you can always remaster the original R3D's anyway later on, so no matter what you know now the flexibility of RAW means it's very unlikely you will have been painted into a corner!.

Graeme - I don't know much about this stuff other than the fact that the math is incredibly difficult and complicated, and even AFTER that there is a major aesthetic/artistic component as well - and that the results you generate are clearly landmark works of technical and creative excellence.

I for one don't question the results and how they were obtained, but do look forward to other people putting together a nice analysis of what exactly this new color science means to the DOPs and Gaffers, while hopefully you are busy with an even greater challenge and changing the game once again!
 
Craig - you hit the nail on the head. It's a new "stock", and it's not up to me to tell anyone how to shoot with the new stock. The important thing to remember, is in any new/old comparison, reset the white balance because the kelvin number that makes white white is (unfortunately) different between the versions as I had to improve the accuracy of that number and how it relates to a color temp meter reading.

Graeme
 
Irri said:
I lit up the green screen with tungsten after doing the tests and took a quick shot. A compositor had a quick look at the two and said he couldn't really tell the difference in terms of pulling a key. It's strange because the blue channel in the screen looked visibly cleaner to the eye. I can send you the tiffs if you want to check for yourself.

There is certainly a difference in noise. Not as much as if you use a 1/2CTB to correct. Maybe half the effect on noise (very rough guess).
Click to expand...


...i suspect its something similar to the furnace tools denoise plugin, particularly as it is in the software not the camera.

If this is the case then it has not much benefit to keying (and could actually be worse) as a compositor would normally have access to these types of tools anyway and would probably prefer the raw plates so they can have more control of the denoising. To be honest if you have to greatly denoise a greenscreen shot in software before keying you are already going to have to work harder in the keyer, it needs to be as clean as poss prior to any processing. If it produced less noise in the raw r3d file then that would benefit the compositor greatly.

it would also mean a lot of people are getting very excited over something that has been possible in post for a good while anyway...hope i'm wrong though!

Can anyone compare the render times between the new colour science and old, regarding the lower noise levels produced, be interesting to see if it is slower. also if someone can test it on just a single frame and get as good results, so that redalert can not do a comparison of previous/next frames...this is how the furnace plugin works (optical flow). so you can only load in a single r3d frame into redalert, preferably with a unique filename to the sequence. although it also has a wavelet based approach which can work on single frames so might not mean anything. could be interseting though.

the other worry is how much softer it is going to get. the guy who posted the headshot blue channels earlier, you have to agree it looks a lot softer with the new less noisy result.

i think it's important to know what we are dealing with here. if it is something similar to the furnace tools or not, I would prefer it to be an option in the transcode, after all the RAW workflow is one of the great benefits and I wouldn't like to think there is pixel munching going on without user control.
 
paul gill said:
...i suspect its something similar to the furnace tools denoise plugin, particularly as it is in the software not the camera.

If this is the case then it has not much benefit to keying (and could actually be worse) as a compositor would normally have access to these types of tools anyway and would probably prefer the raw plates so they can have more control of the denoising. To be honest if you have to greatly denoise a greenscreen shot in software before keying you are already going to have to work harder in the keyer, it needs to be as clean as poss prior to any processing. If it produced less noise in the raw r3d file then that would benefit the compositor greatly.

it would also mean a lot of people are getting very excited over something that has been possible in post for a good while anyway...hope i'm wrong though!

Can anyone compare the render times between the new colour science and old, regarding the lower noise levels produced, be interesting to see if it is slower. also if someone can test it on just a single frame and get as good results, so that redalert can not do a comparison of previous/next frames...this is how the furnace plugin works (optical flow). so you can only load in a single r3d frame into redalert, preferably with a unique filename to the sequence. although it also has a wavelet based approach which can work on single frames so might not mean anything. could be interseting though.

the other worry is how much softer it is going to get. the guy who posted the headshot blue channels earlier, you have to agree it looks a lot softer with the new less noisy result.
Click to expand...

Noise reduction is not part of the new color science and render times are the same.

Jim
 
when should we expect to see this hit the SDK?

...i guess its up to the different software makers to create their updates after you release to the SDK so release times can vary?

I'm on a PC using Vegas 9, so right now the new color science isn't really available to me....am I understanding that right?

jeremiah
 
Jeremiah McLamb said:
when should we expect to see this hit the SDK?

...i guess its up to the different software makers to create their updates after you release to the SDK so release times can vary?

I'm on a PC using Vegas 9, so right now the new color science isn't really available to me....am I understanding that right?

jeremiah
Click to expand...

The render times are the same comparing old vs. new on any given computer or application. It speaks to the fact that there isn't some hidden noise reduction in Graeme's new color science. There is still a selectable NR in RED Alert!, but that has always been there and is separate from the color science.

We are trying to update and release the SDK quickly for both the new color science AND RED Rocket so everyone can update their applications for both at the same time.

Jim
 
Thanks for the feedback.....

I will admit I am looking forward to seeing those HD-SDI outputs. Will I no longer have to explain why the image is tinted green?

When can users expect Apple support through FCP using the new Colour Science? A QT Pro update, etc?

Much thanks,
 
paul gill said:
the other worry is how much softer it is going to get. the guy who posted the headshot blue channels earlier, you have to agree it looks a lot softer with the new less noisy result.
Click to expand...

Bear in mind that the noise will be adding a perceived sharpness that isn't really there (Noise can increase the perceived acutance). So the new image is probably no softer, just less noise.

People still need to just get on with using this new 'stock' and see how it works for them rather than getting all technical. No matter how much info there is out there the best DPs still seem to do their own tests of stock, cameras etc.
 
Can someone help me with question I have. Which is the latest RA is it RA 3.76B, which produced the incredible pictures that Sanjin Jukic posttest of the church, and other things? Or is it Build 20-RA20? If so why are people talking about RA3.6? I’m confused, please help me understand. Thank you. By the way what Graeme Nattress accomplish is just simply incredible.

Humberto Rivera
 
Humberto,

the picture of Karlskirche Wien and some other night shots were from early Fall 2008 in Vienna.

I don't remember RA build from that period but you could find it in the RA build history.

For a purpose of the test were processed in RA B3.60 (with blue channel artifacts)

and RA B20 (a brand new RED magic stock or RED Colour Science by Graeme Nattress).

The Bird Flower pictures are all shot with B20 and processed in RA B20.
 
FYI, when I tested resolution, there is no measured difference between new and old with regards to resolution.

Graeme

Simon Dean said:
Bear in mind that the noise will be adding a perceived sharpness that isn't really there (Noise can increase the perceived acutance). So the new image is probably no softer, just less noise.

People still need to just get on with using this new 'stock' and see how it works for them rather than getting all technical. No matter how much info there is out there the best DPs still seem to do their own tests of stock, cameras etc.
Click to expand...
 
RED Alert! versions

RED Alert! versions

Humberto said:
Can someone help me with question I have. Which is the latest RA is it RA 3.76B, which produced the incredible pictures that Sanjin Jukic posttest of the church, and other things? Or is it Build 20-RA20? If so why are people talking about RA3.6? I’m confused, please help me understand. Thank you. By the way what Graeme Nattress accomplish is just simply incredible.

Humberto Rivera
Click to expand...

RED Alert! V 3.76B was alpha code distributed to the Test Pilots. This has now become beta Build 20 / Version 20.0.1. Visual results should be the same
 
Thank you, Stuart, Graeme, and Sanjin. So RA3.76B is now RA 20.0.1, they are the same thing!

Humberto Rivera
 
Test on B17/B20 and RA v3.60-v20

Test on B17/B20 and RA v3.60-v20

These two days I have been working on a comparison of Build 17 and Build 20, in combination with the RedAlert V3.60 and the RedAlert 20.
Our goal was to find out if and where we can see any improvements.

First of all, if I knew how to upload these images somewhere (tiffs @54Mb each) I would do that, but I think I have to resize them, and then they would probably lose the critical information.
Any suggestions welcome.

This was my situation: I have here 2 Red One cameras, no 5529 and 5645, so very close to each other.
5529 was upgraded to Build 20, 5645 had build 17.

I made a Tungsten-lit setup with a macbeth color checker card, some colorful items, a black background and gray table, lit with one Arri 800 on a approx.45 degrees angle to the right of the camera.

The camera was fitted a 40mm Arri Ultraprime T1.9

We recorded on 16Gb cards with Redcode 36, 4K 16:9 format.
iso was 320, 25 fps and shutter at 180 degrees
We shot without any filter but also with Tiffen CTB 1/4 to see if that makes a cleaner image.

At first we adjusted the light from the Arri-light to 2.8 on the Macbeth card, measured with a Minolta lightmeter in cine mode (24/25 fps).

We made the following exposures by adjusting the iris:

2.8
2.8 with CTB 1/4
1.9
1.9 with CTB 1/4
5.6
5.6 with CTB 1/4

This was done with both cameras, I just put my phone in a corner of the frame in the shots with Build 17, to make a visible difference.

All shots are also with a small white card with details about the camera settings: ISO, iris, ctb.

Knowing that the camera is capturing raw, i didn't bother adjusting the ISO on the camera, but instead chose to change that in the RedAlert version that i would be using later.

Settings in RedAlers, both versions:
Color Space: Rec709
Gamma Space: Rec709
Chroma Denoise: Maximum
Debayer Detail: High
OLPF Compensation: High

Default ISO 320

In the RedAlert I would then white balance on ISO 320 on always the same spot in the second lightest gray area on the card and output a full-res. Tiff (4K)
Then I would raise the ISO just before the zebra would become visible on the whites. After a white balance on that setting I would output a TIFF.

I compared the following:
Build17, rendered with RA V3.60 with Build20, rendered with RA V20
Build17 and Build20, both rendered with RA V3.60
Build17 and Build20, both rendered with RA V20

I used Aperture to combine and compare the images side by side

These are my conclusions:

A) with the given camera settings, the camera is UNDER-exposing the image 2 stops compared to the lightmeter.
This is deducted from the iso-value, which has to be cranked up to 1280 in order to use the full range on the Histogram, without having any zebras.
In order to get a properly lit "negative" or Raw image, it is wise to rate the camera at ISO 80-100 :(
I say this because even the image shot on a full open lens (T1.9) was still 2/3 of a stop under, according to the histogram.
This is something serious to consider (for nightshoots mostly): when you want to shoot with a film-like quality (motion blur similar to a 180 degrees shutter) and on Tungsten, you will find an effective ISO 80-100 !!

B) There is a small difference between the 2 Firmware Builds. I can see the blacks being smoother in B20, mostly because of a little less noise in the Blue channel.
There seems to be less "blodgeness" (is that the word?)

C) The CTB 1/4 does help a bit, again a smoother rendering of most colors.

D) when rendered with the same RA-version, it seems that the B20 gives a slightly higher white level: approx. 0.1-0.2 stop. This could mean that the effective ISO is a fraction higher with B20

A comparison between the similar images on the different Builds with their original RA version (17-v3.60 and 20-v20) as well as a comparison between B17/B20 on V3.60 and B17/B20 on V20 shows that the major steps forward are in this software.

A) RA V20 has a lesser color saturation, less pronounced in greens and reds.

B) The "grain" or noise is more uniform, a lot less Blodgy. It improves almost every dark color and the blacks are now a pleasant "grain", instead of moving dark and blueish patches.

C) Shots that are severely under exposed (in my case I under exposed until an equivalent ISO 5120, which is the T5.6 shot) can still be "usable" when graded with care.

I will try to attach some of the images here at a lower resolution. These are all rendered with RA V20

I hope this is of some help to anyone. At least it made my suspicions clear (about the effective ISO) and it proves that the software can make changes eve with old files from previous Builds.

Rob
 
You must log in or register to reply here.
Back
Top