tests - part 2 - B 30, workflow comparisons, signal chain comparisons + 5D

[Gunleik Groven]

Part-01
Preface

Ok - Geek-alert

This is the followup of the initial render-tests with "old" new color-science (summer 2009) (CPU-render) vs RCX (GPU ??? -render) vs Color (GPU ??? render) vs RCX (rocket render) vs RedRocket (rocket render)

I also want to find out what gives the best image out of Color.

Start off with RAW and RED-Room or start from developed DPXs

I have a hunch that the use of LUTs (and the silly way Color handles LUTS) will be part of the answer...

It's part of the bill to test and match 5D to RED ONE M, so I'll try to check out a 5D for the first time...

Here's the original thread:
http://reduser.net/forum/showthread.php?t=41980

In addition it is a comparison between M RED1 and 5D mk2 (as RED will be used and 5D probably will be part of the arsenal for the production I am testing for)

This second batch of tests is done on a 2008 8-core 2,8 MacPro, OSX 10.5.8 with 16 GB of RAM.

Tools used are:
Color 1.5.1 with the new color plugin:
http://reduser.net/forum/showthread.php?t=41389

RED1 - Mysterium
A variety of Arri MasterPrimes
Schneider ND

Canon EOS5d mk 2 - Current (30 fps) firmware.
Canon 35 mm f1.4 prime

RCX 158
Rocketcine-x 1740

http://reduser.net/forum/showthread.php?t=40067&page=2

Clipfinder 2.5 (BUY IT NOW!!)
http://www.daun.ch/hamingja/

And Monkey Extract (Nice to have on a rainy day!)
http://www.rubbermonkeysoftware.com/

As it is also a test of different workflows, I'll try to document all settings and executions as far as I can manage... -

Remember:
This is not mainly a test of how the RED-ONE - M, performs, but how the different tools we use and workflows we prefer affect the images.

Not every aspect of REDness is relevant for the tests I've done, and sometimes my methodology may be @ fault. Please point out if I've done something plain stupid! -

There won't be 1/3 increment stop exposure tests. But I'll do over and under!
I still think these files may have some interest to the general public. I sure know I'd have loved to lay my hands on them...

I'll be posting RAW files alongside DPXs and image crops, and detailed "how I did it" on every step of the process.
If I am doing something wrong. Please report back, and I'll try to re-do, if the faults are consequential to the production I am testing for.

The shots are not made to be beautiful, but to give information.
If someone should think otherwise: I love working with the RED 1
LOL!

Finally:
A big thanks to Norwegian Broadcast Corp and Hans-Georg Daun for letting me publish this material.

And of course to RED and the support and reactions from them on the tests.

So that's the end of post 1....

 

[Gunleik Groven]

5D vs 5D...

5D vs 5D...

Part-02
5D vs 5D

Ouch.

Undisciplined.
Ranting.

But I cannot help it.

Over the tests I did, I got some weird results out of the 5D...
(and the firmware there is not a beta... - )

So I went out to do a couple of new test:

So here's the first test:
No RED. No scaling.

Just plain old: 5D vs 5D @ 1080

This is the 1st scene:

First: the original files:
RAW still
http://www.byacserv.com/test_02/5d/part_02/IMG_7000.CR2

.mov
http://www.byacserv.com/test_02/5d/part_02/MVI_7001.MOV


I exported a 16 bit tiff with preview (yes, I know. But there's no Photoshop on this computer...) and scaled it down to 1920 x 1280 with "Mitchell" in Shake:

http://www.byacserv.com/test_02/5d/part_02/shake_01.png

And had a look @ what happened on a totally non-testy image:


1. The Brickwall

Crop from original still:

Crop from scaled tiff:

Crop from video:

2. The Windows

Crop from original still:

Crop from scaled tiff:

Crop from video:

3. Odd blue-channel sharpening/edge-peaking and compression

Crop from scaled tiff:

Crop from video:

Finally
The movies seemed soft. So I started to take stills with them to check focus:

Here's a cam-align chart 1080 still from the movie (yup.... I'll get back to this in relation to RED ONE...):

http://www.byacserv.com/test_02/5d/part_02/cam_align_mov.pct

And here's a 1080 scaled in Shake with Mitchell from the full-rez tiff:
http://www.byacserv.com/test_02/5d/part_02/cam_allign_img.png

And here's the cam-original files for these tests:

MOV
http://www.byacserv.com/test_02/5d/part_02/MVI_6986.MOV

CR2
http://www.byacserv.com/test_02/5d/part_02/IMG_6985.CR2


Conclusion?

I think 5D may have many good video-applications.
Detailed landscapes and buildings are not amongst them.

And it's not really soft, but it is a bit sharpened, and a bit of aliasing and a bit of lost detail and a bit of sauce, not Jim-style and indeed very weird.

For close-ups and shallow DOF imagery, I guess it might work quite well, though...

-

Disclaimer:
None of these posts are meant to be Canon-bashing.

 

[Gunleik Groven]

Part-03-a

I was starting on the big batch of tests, but the beta - firmware bug makes the shots more or less unuseable...

My Bad.
It said Beta.

It just looked sooo sexy!

Back to shooting new tests...

Will continue the posts tomorrow...

 

[Ben Brainerd]

Conclusion?

I think 5D may have many good video-applications.
Detailed landscapes and buildings are not amongst them.

And it's not really soft, but it is a bit sharpened, and a bit of aliasing and a bit of lost detail and a bit of sauce, not Jim-style and indeed very weird.

For close-ups and shallow DOF imagery, I guess it might work quite well, though...

None of these posts are meant to be Canon-bashing.

One thing to remember with the 5D is that you can change the level of Sharpening and whatnot in the picture-styles menu. Stu Maschwitz discusses flattening out the 5D video image for better detail and a flatter gradeable image here: http://prolost.com/blog/2009/8/3/flatten-your-5d.html

That said, the 5D and any of the REDs are completely different beasts. 5Ds/7Ds absolutely suck if you point them at a resolution chart. They don't shoot *good* video. Most of us (Yeah, I'm a 7D shooter) are very well aware of that. What they do is shoot *pretty* video. To quote Stu: “Buttons and features and resolution charts just had their asses handed to them by sex appeal.”

RED is *always* going to kick a 5D/7D's ass.

Sorry, minor rant over. Back to your regularly scheduled camera tests. Don't want to threadjack.

 

[Gunleik Groven]

One thing to remember with the 5D is that you can change the level of Sharpening and whatnot in the picture-styles menu. Stu Maschwitz discusses flattening out the 5D video image for better detail and a flatter gradeable image here: http://prolost.com/blog/2009/8/3/flatten-your-5d.html


Sorry, minor rant over. Back to your regularly scheduled camera tests. Don't want to threadjack.

Thanks for the input.
Makes total sense.

Actually my plan is to make the 5D work together with the RED, I just need to look a bit under the hood to see what's going on. It's not really intended to be a Canon rant. (even though I like to peek under the hood...)

 

[Uli Plank]

The reason is simple:

The 5D is an excellent still camera, and it's OLPF is optimized for stills. It can't be optimized for video at the same time, since other than the R1 it can't read all the pixels at film speed. The pixels get binned, not scaled. So you get massive aliasing!

I consider it's true resolution no more than 720, the rest is fake (aliased) detail.

The other problem is H.264 with 8 bits and 4:2:0 subsampling. Not that great for any grading. If you use a flat profile (as suggested by some) you run into banding if you want some contrast back.

Admitted, it's images can look sexy at first view. The RED's look sexy at second and third look and after grading too…

 

[Gunleik Groven]

Stress-test 01

Stress-test 01

Just a quickie today
(had to re-shoot the tests and have to re-write the text...)

What's the best current way of bringing an underexposed image into a space where it is actually gradeable without adding too much noise.

Here's a seriously underexposed Chroma de monde chart (R3D) Light was daylight + 0.6 ND @ T11 on a Arri MasterPrime/25mm.

http://www.byacserv.com/test_02/red/under/B003_C011_0316WP_001.R3D

Looks a bit like this "as shot" @ 320/5600k:

After a bit of treatment, the image looks like this:

Thing is, I developed the image two ways (in Clipfinder 2.5 with new-new color science)



First I used these settings:

And I got this result (blue-channel crop...)

The full 4k TIFF is here:
http://www.byacserv.com/test_02/red/under/B003_C011_0316WP-000005.tiff

Then I used these settings:

And had this result (blue-channel crop):

Here's the full 4k tiff:

http://www.byacserv.com/test_02/red/under/B003_C011_0316WP-000016.tiff

I think I'll try to develop these with the old-new color science, too, but the results are quite impressive for the new-new
I am particularily impressed with how the resolution basically comes through, even with a stress test like this. (Have a look at the trumpets in the 4k developed files...)

Old truths are still true and the sensor is still the same:
Dragging info out of the lowlights with any kind of gain function (like ISO/exposure/FLUT™) creates noise, which will make the image harder to grade.

Forcing the image out through curve/brightness & contrast + even lowering the exposure/ISO/FLUT™™/gain-controlls, will give you an image with less noise for grade.

I didn't change the denoise settings from the default for this test.

If you look at the tiff, you'll see that the noise difference is quite significant in all channels.


RED-1 M
B30.xxx

Cheers!

 

[Gunleik Groven]

Prt 03 - Scaling to 1080. 4k HD or 4k 16:9?

Prt 03 - Scaling to 1080. 4k HD or 4k 16:9?

Part-03 - scaling to 1080

RedRocket vs Redline vs Color - scaling tests

4kHD vs 4k 16:9

Here's a zoneplate

@ once: It's not perfectly alligned, but it'll do for this test...

http://www.byacserv.com/test_02/to_1080/B003_C005_0316FG_001.R3D

What I wanted to find out is what app scales the best to 1080, and how much better does 4kHD skale to 1080 than 4k 16:9

I didn't shoot 4kHD... -
But I scaled 4k 16:9 to 2k & 1080
The 2k scales, should be the same as a 4kHD scaled to 1080.


Clipfinder 4k 16:9 to 2k 16:9 with these render-settings (similar scaling to 4k HD to 1080)

the dpx is here:
http://www.byacserv.com/test_02/to_1080/cf_4k-2k.dpx

And: here's a crop:

RedRocket 4k 16:9 to 2k 16:9 (similar scaling to 4k HD to 1080)

Here's the DPX:
http://www.byacserv.com/test_02/to_1080/rr_4k-2k.dpx

And here's a crop:

Color 4k 16:9 to 2k 16:9 (similar scaling to 4k HD to 1080)

Here's the DPX:
http://www.byacserv.com/test_02/to_1080/color_4k-2k.dpx

And here's a crop:

Clipfinder 4k 16:9 to 1080

The render-settings:

The DPX
http://www.byacserv.com/test_02/to_1080/cf_4k-1080.dpx

The crop:

RedRocket 4k 16:9 to 1080

The DPX
http://www.byacserv.com/test_02/to_1080/rr_4k-1080.dpx

The crop:

Part-03 - scaling to 1080

RedRocket vs Redline

4kHD vs 4k 16:9

Here's a zoneplate

@ once: It's not perfectly alligned, but it'll do for this test...

http://www.byacserv.com/test_02/to_1080/B003_C005_0316FG_001.R3D

What I wanted to find out is what app scales the best to 1080, and how much better does 4kHD skale to 1080 than 4k 16:9

I didn't shoot 4kHD... -
But I scaled 4k 16:9 to 2k & 1080
The 2k scales, should be the same as a 4kHD scaled to 1080.


Clipfinder 4k 16:9 to 2k 16:9 with these render-settings (similar scaling to 4k HD to 1080)

the dpx is here:
http://www.byacserv.com/test_02/to_1080/cf_4k-2k.dpx

And: here's a crop:

RedRocket 4k 16:9 to 2k 16:9 (similar scaling to 4k HD to 1080)

Here's the DPX:
http://www.byacserv.com/test_02/to_1080/rr_4k-2k.dpx

And here's a crop:

Color 4k 16:9 to 2k 16:9 (similar scaling to 4k HD to 1080)

Here's the DPX:
http://www.byacserv.com/test_02/to_1080/color_4k-2k.dpx

And here's a crop:

Clipfinder 4k 16:9 to 1080

The render-settings:

The DPX
http://www.byacserv.com/test_02/to_1080/cf_4k-1080.dpx

The crop:

Color 4k 16:9 to 1080

The DPX
http://www.byacserv.com/test_02/to_1080/color_4k-1080.dpx

The crop:

For reference - Canon 5d
Full chart
http://www.byacserv.com/test_02/to_1080/canon_cam_align_mov.png

The Crop:

What can I say?
Other than that the RedRocket does an incredible scaling job...
And Color is surprisingly uniform with the Rocket (not identical, but very similar except for the expected levels differences...)

Out of these first tests, Color seems like an unbeatable value for money.
But hang on...
I'll have to look a bit into the Color signal-chain and distortions to make a verdict...

Some of the problems with this redline render, probably is due to the Mitchel Scale?

With a RedRocket installed, I really see no reason NOT to shoot 4K 16:9 for 1080 delivery.

(but YMMV...)

Cheers!

Gunleik

 

[Gunleik Groven]

I'll be pushing out as many tests tomorrow as I manage, but if you guys would like them, I could assemble a .zip file with a bunch of RAW testfiles for others to analyse their own workflows with.

Next up is a bit of color signal-chain testing. (better results from RAW or DPXs, with and without LUTs involved).

BUT I have shot quite a bit of material that I will use for analysis after easter.

I could upload them (or a selection of them) for you guys to run your own tests.

Also, I would be interested in seing results done in a similar way from Scratch, Speedgrade, Smoke and other RAW tools (that would be similar AB tests to the ones I'm gonna do with Color).

Anyho'

Till tomorrow...

G

 

[Ryan E. Walters]

Wow, very impressive results. Thanks for posting these,I look forward to following the results of the rest of the tests.

 

[Gunleik Groven]

Quickie...

Updated the scaling test with Color results.

Color begins to look like an incredible little app, after being quite lacking since the first RAW plugin landed...

http://reduser.net/forum/showpost.php?p=569198&postcount=8

 

[Gunleik Groven]

Quickie two...

I don't have time to post all the results today, and it will thus take some time before these threads update (around april 6th. in fact).

Sorry for that.

I've been playing around with color in DPX and RAW today.

And even though there are still issues that I find to be significant glitches with the Color implementation of the RED room, it's become an incredible little app.

I won't conclude on anything before I am through, but I am really surprised how much the new beta-plugin improves the RAW workflow and image-generation in Color.

My hat is off.

So, I have a question:

Have anyone done longform (+20 minutes acts) RAW grades in Color without major (bug) issues like:

These are bugs I have seen (but haven't tested the update for as I am particularily checking signal-chains)

Conform/TC - related
Grades of particular shots refusing to render right?
Making the secondaries work as you would expect?
(The color-picker seems to pick from the camera-metadata image and not the "after the red-room" image.)
Other randomnesses particular for RAW.


The minor issue is the lack of controll in the RED-Room.
FLUT... I don't miss so much as long as it works it magic with the highlights, which it seems to do pretty well under the hood, without the slider.
I miss the curve-tool the most...

For tricky high-dynamic and low-lit shots, I can develop them better in some of the other tools.
But that's like for a very rare amount of shots I'll have to do tweaking like that.


The big issue I see out of the box is the levels limits, especially - but not only - in tandem with how Color uses "display LUTs" and the Gamma issues.

(try playing the same image over SDI to an external monitor/waveform off FCP and Color, and you'll see what I mean...)
The setup in FCP and Color differ.
Even over SDI.

One could compensate for this with a LUT, but...

The LUT is put first in the signal-chain, like a grade, and when one takes it off to render, one risk to go above or bellow the "broadcast safe" limits.

This is a bit like the
"Can we please meter RAW when we shoot" issue on the cam itself...
Just silly - untill it one day is incredibly fixed. (Yes. I am a big B30 fan!)

The other big issue is related to the same thing.
With log DPXs you can have levels above and below "broadcastsafe (0 & 100), that you can recover or push down in the secondaries.

The way color works now, it is harder/impossible to recover darks and highlights out of the primary into the next rooms. With dpxs this is not an issue.

Except for these?

Fill me in.

Cheers!

Gunleik

 

[Ryan E. Walters]

Is there any chance that you can shoot the zone plate again in 4k HD and then scale it to 1080p? It was my understanding that shooting at that resolution will yield a better down sample to 1080p due to the even integers. Not shooting at that resolution for the test makes it not a true comparison ...

Thanks for all of your work tho- it is very helpful, and interesting to see the results. I have stopped using Color due to it's past problems with scaling (everything was always soft), and the lack of user controls in the RAW. It looks like they are getting better on the scaling issue- but the lack of user controls still frustrates me ... hopefully that will come, along with support for the Red Rocket.

 

[Uli Plank]

Making the secondaries work as you would expect?
(The color-picker seems to pick from the camera-metadata image and not the "after the red-room" image.)

The keyer in the secondaries is reverting to the incoming data without any primary correction – not only with R3Ds, but it's more obvious there. It's currently being discussed at length at the Cow's RED forum.

But you can always mimick those functions in the FX room, which is working downstream as expected.

 

[Luca Immesi]

The keyer in the secondaries is reverting to the incoming data without any primary correction – not only with R3Ds, but it's more obvious there. It's currently being discussed at length at the Cow's RED forum.

But you can always mimick those functions in the FX room, which is working downstream as expected.

Thanks Uli, could you explain more about mimicking the secondaries in the FX room?

 

[Gunleik Groven]

Is there any chance that you can shoot the zone plate again in 4k HD and then scale it to 1080p? It was my understanding that shooting at that resolution will yield a better down sample to 1080p due to the even integers. Not shooting at that resolution for the test makes it not a true comparison ...

Thanks for all of your work tho- it is very helpful, and interesting to see the results. I have stopped using Color due to it's past problems with scaling (everything was always soft), and the lack of user controls in the RAW. It looks like they are getting better on the scaling issue- but the lack of user controls still frustrates me ... hopefully that will come, along with support for the RED Rocket™.

As to scaling

4k 16:9 to 2k 16:9 should in theory give exactly the same results as 4khd to 1080 (to go to 1080 from 2k, you just crop off the same pixels as the cam does. The results should be equal...)

 

[Gunleik Groven]

Thanks Uli, could you explain more about mimicking the secondaries in the FX room?

Uli!
That was a good idea.
Didn't think of that as an option.

Luca.

You can use the keyer in the FX room to add keys for selected "colors and values" similarily to the secondaries.

Crunching my head a bit around multiple secondaries, but I can see how that could work well.

Actually one COULD use Color a bit like one had to with the Scratch I tried - back in the day when Itried it (dunno if that's how it still works)...

Using the RedRoom and primary for a "pre-grade" and then do the finnishing in the FX and primary out...

 

[Ryan E. Walters]

As to scaling

4k 16:9 to 2k 16:9 should in theory give exactly the same results as 4khd to 1080 (to go to 1080 from 2k, you just crop off the same pixels as the cam does. The results should be equal...)

Hmmm, I'm not 100% sure that is correct- in one instance (4kHD) you are scaling down to 1080p, and in the other instance you are cropping (2k 16:9) to 1080p. So while the end result is theoretically the same, there is potential for the actual results for be different in practice. The 4kHD frame has more information in it, which should yield a better result - at least that is my guess anyway.

 

[Luca Immesi]

As to scaling

4k 16:9 to 2k 16:9 should in theory give exactly the same results as 4khd to 1080 (to go to 1080 from 2k, you just crop off the same pixels as the cam does. The results should be equal...)

Thanks Gunleik, I was thinking about multiple secondaries.

 

[Gunleik Groven]

Hmmm, I'm not 100% sure that is correct- in one instance (4kHD) you are scaling down to 1080p, and in the other instance you are cropping (2k 16:9) to 1080p. So while the end result is theoretically the same, there is potential for the actual results for be different in practice. The 4kHD frame has more information in it, which should yield a better result - at least that is my guess anyway.

Ok... I'll shoot a 4khd chart over easter...