Welcome to our community

Be a part of something great, join today!

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 :)

OK - the RCX vs Color vs Redline/Clipfinder debayer noise-test when preparing images

I think we're miscommunicating a bit.

If you took this image and exported a 2k and a 4k dpx with redlog settings and everything else flat, and took the same image and did the same exports from RA/clipfinder, and then opened the resulting dpxs in Shake and peeked into the red, green and blue channels (not the whole image) - I bet you'll find basically the same differences I have.

I'll even take your R3D and check, if you ftp it to me.

Anyway.

We are seeking different information obviously

I learned that the debayer in Color and RCX is close to identical.
I would never have put my money on that in advance.

I also learned a couple of other things.

I am happy!

Next week I'll shoot some B-30 material on MP at different apertures and run through even more extensive tests, and I'll hopefully be able to post the R3D's with them for others to confirm, or find me wrong.

I don't need to be right at all times...

Cheers and peace!

Gunleik
 
These tests are really great! Thank you so much for doing this so scientifically! Have we come to any definitive conclusions yet? I'm particularly interested in the sharpness results using 4K R3Ds in Color, downsampling to 1080p, V.S. doing so BEFORE Color in REDCine-X via REDRocket. Thanks advance for sharing your thoughts...

Best,

- Jordan
 
I don't think there are necessarily any significant "conclusions" to come to, Jordan. If anything, Gunleik has managed to myth-bust a number of discussions, though.

For example:

1. "RedCine-X produces a less-noisy image than Color." BUSTED. I'll post images if people are really too lazy to do the check themselves, but take the images Gunleik posted most recently (the 4K DPXs), load them into Photoshop and put an auto-levels on.

Look at the sky. There is *NO* difference whatsoever.

2. "RedCine-X produces a *sharper* image than Color." I think people are confusing what shows up in Color's preview window with it's actual output. (Now, the fact that the preview isn't a 100% faithful representation is a whole other story...) But looking at DPX output from the two programs, I can't see any difference, sharpness-wise.

Now, sure, RC-X *has* sharpness controls you can use. But there's no difference in the baseline tools. So sharpness is a matter of taste, and where you feel like putting it in your workflow. Like it should be.

End of mythbusting.

Personally, I'm pretty shocked that Clipfinder is consistently producing the least-noisy images. I figured the results in the first test were a side-effect of the scaling method. But even at 4K, Clipfinder is producing noticeably cleaner imagery. There seems to be a bit of color-shift with CF, though, so I'm guessing it's not using the latest and greatest FLUT™ science.

Like Jordan, I would like to see some scaling tests, if someone's up for it. I don't have any 4K footage at the moment, or I'd do it myself.

In particular, I'd like to see:
1. RC-X 4k (Full Debayer) to 1080p
2. RC-X 4k (Half/Half-High) to 1080p
3. Color 4k (Full) to 1080p
4. Color 4k (Half) to 1080p
5. Clipfinder 4k (Full) to 1080p (Using the different scaling methods, since CF appears to be the only one that lets you choose your scaling method anymore)
6. Clipfinder 4k (Half/Half-High) to 1080p

I've actually been pondering where the best place to scale is for a while. Would be nice to get some definitive answers there. Or at least some good info.

**EDIT**
Ok, so it turns out I *do* have some 4K footage. I'll do these scaling tests myself, though if someone else wants to do them as well to verify, be my guest.
 
This is def info gathering, more than conclusion bashing.

That said... -:)

AFAIK neither RCX nor Color use the cpu/redline for rendering, but rather the GPU.
This means that it is not impossible that different GPUs will give you different results...

Actually the differnce between gpu and cpu renders look somewhat familiar to me from some tests i did a long while back. GPU renders back in the old days were consistently less consistent. I am not a computer tech, so I have no idea why.
these results - at least - are consistent.

nuff said. color and rcx most likely uses another method for debayer and render than cf.
Same SDK, probably. Different method.

i think there are both more details and less noise in the redline/cf renders.
Is that sharper?
Define sharper!
There's more info and a better startingpoint for a grade (I think)
The colorshift comes from the difference between CamRGB and RedColor - I guess.
I just tried to set optimal settings in each app, without crushing blacks in a way that would disguise differences in the lowlight renderitions.

but I saw something in rcx/color that sorta looked like an auto-drx function when the highlights blow. not sure exactly what, but different. In 4k, they went smoother into white than Redline. That is nice.

I am really curious on how redrocket renders these files, and I'll know tomorrow. Probably won't post the results until Sunday, though.

As to scaling tests, I'll do them end of next week and the beginning of the week after with fresh B30 shots of charts and gradients. I'll post results if I am allowed to... I will give priority to the tests I need most for my own decissions, though, but I hope to upload small r3d's for you all to test with.

But my conclusion is that for critical work, there's nothing that beats redline/clipfinder...

Yet...

A. least until I have tested the Rocket

But that's still just my conclusion thus far. I need to do some more tests to decide. And I'll continue to render stuff directly from cam-proxies in FCP, if that's what is right for the job.

I agree with Sanjin that on the long run RCX is probably the way to go. My tests are more aimed at them who have projects to finish today...
 
Gunleik Groven said:
Incidentially I think I the the same issues in your post, even though it's hard to tell for sure because of the jpg-compression/8bit/low rez it has.

You judge.

A blue channel crop from your Mercedes image:
http://www.byacserv.com/noise/4k_and_2k_5000k_and_2000k/peek_4k_2k/merc_blue_crop.png

I am not bashing RCX or R1, just merely testing workflows and codec.
Click to expand...

Gunleik,

I continue to show you what I see and what I'm talking about.

Here are the next examples {shot handheld on R1 with Nokton 50mm f/1.1 at wide open @ f/1.1 [extremely short DoF], 2K output from RCX, REDColor and REDLog}:

DPX_DJV_04.jpg

Here is a collage from DJV Sourceforge DPX player and analyzer (free app) that can btw, play DPX 2K on a normal desktop realtime.

DCX_Blue_01.jpg

Blue channel derived from tiff 16 bit output from DCX.

DCX_Blue_02.jpg

Blue channel in monochrome derived from tiff 16 bit output from DCX.

I can't upload via ftp because there are monster sized big files but can offer you two files

one dpx and another tiff 16 bit and you can have a look at them yourself.

DPX 2K>>>

TIFF 2K 16bit>>>
 
Thanks Sanjin. I'll have a look tomorrow, so please let the files stay up untill tomorrow evening.

But you're rendering with the Rocket aren't you?
That's what I am testing tomorrow.

If RR renders better than RCX without Rocket, it would be great!

And that would give RCX the edge over Color as an instrument for a technical pass before grade.

So many girls, so little time...
 
Since clipfinder uses redline, it would seem the only difference is that RCX is using the new color science vs the old. In my experience, the rcx images are better for grading and color and overall quality which to me is more important than pixel peeping the blue channel.
 
Gunleik Groven said:
but I saw something in rcx/color that sorta looked like an auto-drx function when the highlights blow. not sure exactly what, but different. In 4k, they went smoother into white than Redline. That is nice.
Click to expand...

I'm pretty sure that's FLUT, which the new Color plugin supports, but doesn't have a control for. One of the things FLUT is supposed to do is keep highlights from clipping. Compress the hell out of, sure, but they're theoretically not clipping. That's probably what you're seeing there, and it's pretty awesome.

I'm currently uploading the results of my basic scaling test. Nothing fancy, just grabbed a pair of 4K clips and ran them out through the various available 1080 outputs. Clipfinder, using every last one of the scaling filters, RC-X and Color.

To be kinda mysterious like Gunleik, some interesting results. :) Actually, I got amazingly bored with all the pixel-peeping. More when I get the files/links up.
 
Carsten Fenstermacher said:
In my experience, the rcx images are better for grading and color and overall quality which to me is more important than pixel peeping the blue channel.
Click to expand...

But without this pixel peeping, it's impossible to say *why* RC-X looks better. And to be honest, the more I'm looking at it, the more I find that, aside from the odd gamma shift, RC-X looks no different from Color.

Also: Sanjin, maybe it's the fact that I'm really tired, but I'm not getting the point you're trying to make here. I'm seeing a lot of images, but not much in the way of comparison with other tools.

I'm getting a "RedCine-X is producing better results" vibe, but again, I'm not entirely sure where you're going with it.
 
Ok, time for my massive contribution to the discussion: Scaling! (Yay?)

Some notes:
1. I tried to pick some footage/frames that have some good lines for checking detail against. In the 01 clips (Dirt Road), I suggest the concrete blocks/rebar on the left, and the mailbox on the right. For 02 (The parking lot) I suggest the fence in the background and the cars/license plates.

2. This is all build 17 footage, run out of the various programs at the metadata settings. ISO 320, RedSpace/RedSpace. I didn't think about moving everything to RedLOG until about 5 minutes ago, when I was already almost done uploading.

3. Bunch of DPX files here, since it seemed like the easiest way to pull the stills and maintain the quality necessary to peep at them. I'll rip off some PNGs or JPGs if people really want them.

Files:
Clip 01 - Dirt Road:

4K DPX (From Clipfinder)
http://www.ragingmushroom.com/RED_tests/01_4K.dpx

1080 Scales from Clipfinder:
Bell:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Bell.dpx

Catmul/Rom:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Catmul.dpx

Cubic B-Spline:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Cubic.dpx

Gauss:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Gauss.dpx

Lanczos:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Lanczos.dpx

Mitchell:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Mitchell.dpx

Quadratic:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Quad.dpx

Wide Gauss:
http://www.ragingmushroom.com/RED_tests/01_CF_1080_Wide_Gauss.dpx

1080 Scale from Color:
http://www.ragingmushroom.com/RED_tests/01_Color_1080.dpx

1080 Scale from RC-X:
http://www.ragingmushroom.com/RED_tests/01_RCX_1080.dpx


Clip 02 - Parking Lot:

4K DPX (From Clipfinder)
http://www.ragingmushroom.com/RED_tests/02_4K.dpx

1080 Scales from Clipfinder:
Bell:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Bell.dpx

Catmul/Rom:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Catmul.dpx

Cubic B-Spline:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Cubic.dpx

Gauss:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Gauss.dpx

Lanczos:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Lanczos.dpx

Mitchell:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Mitchell.dpx

Quadratic:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Quad.dpx

Wide Gauss:
http://www.ragingmushroom.com/RED_tests/02_CF_1080_Wide_Gauss.dpx

1080 Scale from Color:
http://www.ragingmushroom.com/RED_tests/02_Color_1080.dpx

1080 Scale from RC-X:
http://www.ragingmushroom.com/RED_tests/02_RCX_1080.dpx


My thoughts:
1. To be perfectly honest, I can't tell the difference between the various CF scaling methods until I zoom in to about 400% in Photoshop. And even then the differences are minute. If you'd rather not download all of the DPXs, check out Lanczos, Mitchell and Catmul/Rom, which seem to produce the best results to my eyes. (Lanczos/Catmul = Sharpest, Mitchell = Smoothest)

2. Again, aside from the gamma shift, Color and RC-X seem almost indistinguishable. I was actually surprised by this, since this is one area where they're *not* drawing on the same tools (RED SDK). I guess they're both using the same, or similar, scaling algorithms.

Results:
Buggered if I know. :P I was hoping for something dramatic, that would tell me whether it was best to grade then scale, or scale then grade. Personally, aside from space considerations, I think the issue is kind of moot.

I don't think my workflow will be changing any, as to my mind, grading can change the level of perceptual detail, so it's better to grade at full resolution, then scale after the fact. Maybe someone else can convince me otherwise. There's also the fact that, as I don't own a Rocket, rendering any significant amount of footage to DPX or scaling in advance requires a butt-load of time.
 
Gunleik Groven said:
Thanks Sanjin. I'll have a look tomorrow, so please let the files stay up untill tomorrow evening.

But you're rendering with the Rocket aren't you?
That's what I am testing tomorrow.

If RR renders better than RCX without Rocket, it would be great!

And that would give RCX the edge over Color as an instrument for a technical pass before grade.

So many girls, so little time...
Click to expand...

Gunleik,

yes I'm on RR pretty long (relatively) and it is must have with RED1 M or MX, later Epic and Scarlet.
 
Sanjin

Here's a similar test with your 2k dpx.

To me it seems pretty much in sync with what I found myself when rendering out of Color/RCX, but you judge!

http://www.byacserv.com/noise/peek/sanjin_blue.png

the only way for me to tell if there was a difference was if you made a dpx with clipfinder from the same frame with these settings:

camrgb
redlog
320 asa
exp = - 0,4
Resize = mitchel
2k
no denoise or sharpening algorithms
No DRX
5600k

If you give me that dpx, too - it is possible to compare (given that the one I got was developed with the same settings I have described above...)
 
In Gunleik tests it seems a mixed bag. The girl looks better in redcinex overall and in blue but the room scene looks better in redcinex overall but not for the blue channel.

However, I've found I get fantastic results in redcinex when i white balance in redcineex so maybe that's where Gunleiks testing is missing a step?
 
Mike Gross said:
In Gunleik tests it seems a mixed bag. The girl looks better in redcinex overall and in blue but the room scene looks better in redcinex overall but not for the blue channel.

However, I've found I get fantastic results in redcinex when i white balance in redcineex so maybe that's where Gunleiks testing is missing a step?
Click to expand...

I did some ballancing tests in the 4k/2k @ 5,6k/2k kelvin to check for both scaling and WB issues.

But I'll do some B30 shots to verify.
 
Mike Gross said:
In Gunleik tests it seems a mixed bag. The girl looks better in redcinex overall and in blue but the room scene looks better in redcinex overall but not for the blue channel.

However, I've found I get fantastic results in redcinex when i white balance in redcineex so maybe that's where Gunleiks testing is missing a step?
Click to expand...

Just to reiterate as well, we're not trying to prove that one particular program is "better" than another here. We're just doing some comparison and contrasting.

At most, we're proving that some of the tools, like the much maligned Color plugin, are better than people give them credit for. Does RC-X produce fantastic images? Yes. Does Color produce fantastic images? Also yes. Does Clipfinder? Surprisingly, yes, considering it's not running the latest color-science iteration.

One thing to consider, though, is that we have to *work* with these tools. Pixel-Peeping tests are all well and good, but at some point we have to buckle down and get work done. And when it comes time to actually work... I've got a nice little rant about programs that are actually usable vs. programs that look flashy but give me headaches, but I'm thinking this isn't the thread for it. Not entirely sure where to put it, but I've gotta get it off my chest somewhere...
 
You must log in or register to reply here.
Back
Top