RED vs. Film - A Practical Comparison

[Sanjin Jukic]

This is one of the latest of tests with RED1 vs. Film that was not discussed here.

I found it recently on CML (Cinematography Mailing List).

The cinematographers Alex Michael Smith, Benjamin Verhulst that did this test are mainly talking about some surprising results here and there (the RED rolling shutter issue)...

The test is made at THE FLORIDA STATE UNIVERSITY COLLEGE OF MOTION PICTURE ARTS

and it was "a thesis submitted to the College of Motion Picture Arts in partial fulfillment of the requirements for graduation with Honors in the Major".

Here is the LINK of this test>>>

Let me know what you think?

 

[Bryan Arnold]

Nice comparison, the Red One has the best resolution in this test. I was surprised to see that 16mm had better DR than the Red One and 35mm. What's up with that? I also didn't knew that the roling shutter was that bad on the Red One.

Seems like the same end result can be achieved with both formats if a proper workflow is taken into consideration.

 

[MichaelP]

I am wondering whether he meant to say HDCAM SR instead of HDCAM keeping it RGB throughout the process as with the RED and 35mm.

Michael

 

[Mitch Gross]

This test is flawed in so many ways it's hard to count.

 

[Ryan E. Walters]

Not a valid test.

Not a valid test.

Also pointed out in the CML discussion is that this test is not valid for many of the procedural flaws in the test. At a bare minimum, the fact that the 16mm looks better then the 35mm should throw up a red flag.

I'm really interested in seeing a well done test like this, but if the ASC & BSC had such a difficult time doing it right ...

 

[David Mullen ASC]

I don't get why the film was more underexposed than the RED in some shots and more overexposed in other shots. If the difference was consistent, at least I could put that down to the rating of the stock or cameras.

Also, the transfer of the 35mm film footage was a bit ugly in terms of the whites and blacks.

I know these tests are hard to do -- I'm sure if I did one there would be testing consistency problems that would come up -- so I'm trying to not get too critical. I'm just curious about some of the elements of the test.

I'm not too surprised that the resolution of the RED was comparable to the 35mm, but I wonder if the jitteriness of the 35mm is contributing a bit to the softness, which makes me wonder about the camera's registration and the steadiness of the transfer device, though I obviously expect the RED frame to be steadier.

It's the exposure differences that don't make much sense to me. Since the film negative should have had at least another two stops of dynamic range, where did it go?

But it speaks to the basic thing that always comes up with these tests -- when the results conform to your expectation, you like them, and when they don't, they bother you.

 

[Patrick Tresch]

Strange test.

Strange that they don't mention in the introduction that R3D is a compressed RAW format.
Strange Telecinema/scan they did.
Strange that 16mm and 35mm don't have the same exposure/density when comparing latitude...
Strange they can compare resolution of 35mm/16mm/RED when going to a 2k QT...

That said... it's incredible that film/fotochemical processing and handling is getting poorer as digital/DI is getting better.

For exemple a 16mm optical blow-up made 10 years ago (by Schwarz film in Switzerland) beats any 16mm 2k scan/DI/arrilaser "blow-up" made today.

The only conclusion I can agree with them is :

"If digital projection becomes the norm, will the increased resolution of film even matter if it is not transferred into the digital domain?"

Patrick

 

[Sanjin Jukic]

Obviously that this test is done by the TWO STUDENTS...

...probably with a lot of mistakes included...

...but anyway it's good to know what is going on in these type of cases that could actually happened everywhere....

...even in Switzerland, Russia, Japan, USA, South Africa, Brazil, Singapore, Thailand or Austria...

...mostly I'm learning from these sort of efforts or mistakes, somebody else's or mine...

 

[Benjamin Verhulst]

We've found there should be a prelude going into the paper, this is a slightly modified version of what I just posted to the cml list:

Thanks for taking it easy on us
I thought I had made it clear this was an honors thesis, I don't want to claim to be something we're not.

The tests were designed to mimic what actually happens in the real world, not to compare the absolute best scenarios for each format.
Hence the 35mm was on a 4k scan downres'd to 2k (10-bit dpx) (when was the last feature you worked on that scanned the entire thing at 6k downres'd to 4?) and the S16 was telecine'd (for comparison to what many low-budget films and most film schools work with - one-light to charts).

And yes, there were definitely some constraints with donated services, etc. That being said, I think we're doing something you don't see that often - comparing formats in ways they're actually used.

The image sizes are different because the 35mm was full-frame academy and the RED is the equivalent of Super35 (yes, this affects the resolution to some extent). The exact same lens was used on the 35 and RED at the same stop and focus with the exception of strobing tests when the cameras were rolling at the same time. The Super16 was using a lens from the same set with an equivalent field of view (by 'matched' I mean all lenses were re-glassed at the same time).

We never shot with any glass in front of the lens and the RED was 'rated to 3200' in the metadata. We used 500T because that's what most people would shoot in this situation. Again, we aimed to compare real life scenarios, not 'best for the camera' scenarios.

The workflow was designed to be as pure as possible - the R3Ds were exported from RED Cine using the RGB LUT standard as a quicktime 10-bit uncompressed 4:4:4 file. The resulting files were minimally adjusted in Color to match charts, but there were no LUTs applied after being exported from R3D.

We were confused by the lack of latitude in the 35mm scan as well, but the working print looked fine, great even (again, one-light to charts). Unfortunately I don't have a way to show you that, but you do get a little bit more exposure latitude when the dpx is graded. Ideally, we'd really like to get another scan, but didn't want to push the issue as it was donated.

Our next step would be to grade the images from the 'raw files' and compare them, but we're having trouble thinking of a way to do so fairly. So much of what the 'best look' for an image is personal preference.

Thanks for the critique, it's what we were hoping for. In keeping with the spirit of comparing these formats in ways they are actually used by most of the film world and not just in ideal situations, how would you design the post workflow? We're willing to go back and re-work from the original files and even make them available online if we can find a way.

Benjamin Verhulst
benjaminverhulst.com

 

[Patrick Tresch]

Thanks Benjamin,

Hope I didn't offend you or minimize your work. Have a nice new year.

Patrick

 

[Patrick Tresch]

BTW beautiful picture!

 

[David Mullen ASC]

I know these tests are hard, especially with donated equipment, time, etc.

One question I have is the 35mm camera -- it seems awfully jittery if that's a pin-registered scan at 4K. I know it's being way magnified, but I was wondering what camera was used and if you did a registration test before you shot the comparison test.

Steadiness and focus can affect a resolution chart test. Also, the downsampling from 4K to 2K can sometimes affect it. Was the RED footage also downsampled to 2K using the same process?

I think most of us film shooters are mainly surprised at the compressed highlight detail in the negative scan, it doesn't quite make sense unless the film stock was way overexposed, misprocessed, or scanned improperly. In fact, I'm somewhat surprised that your HDCAM transfer of the Super-16 has more highlight detail than your 35mm scan. Generally it's in the white areas that I have the most problem with using HD as an intermediate for film. Yet in this case, it's your 35mm scan that has more noise in the whites.

Did you list all the tech parameters somewhere (camera, lens, film stock, type of scanner, etc.?) I must have skipped past it in the article.

Also, if you shot 35mm Academy, not Full Aperture, wouldn't the scan not be 4096 across but something in the 3.7K-ish area? And I assume you shot the RED at 4.5K if you managed to get a Full Aperture (24mm) width image.

 

[Benjamin Verhulst]

It is a bit jittery and very zoomed in. The 35 stock (5219) was shot on a BL4 with Zeiss Super Speeds, and no, we did not do a registration test. I believe it was scanned on an Arriscan. The down-sampling occurred at the lab in the process of creating the dpx from the raw data.

By Full Frame Academy, I mean 35mm Full Academy (1.37), sorry for any confusion. We were framing for 1.85. I just put a sample dpx online and will throw some others on later. It's full 2k (2048x1556), so I can't be sure if it the scan was 4096 or slightly less. I'll inquire with Fotokem.

http://www.benjaminverhulst.com/dop/RED_vs_Film_files/rvf_test_flt_txtd_111009.0002012.dpx

We expected to have to compensate for exposing the 5219 at 320 ISO, but there is really very little left in the highlights, which threw us in a spin. We triple checked and everything had been exposed properly. Again, the working print looked fine and had all the detail we'd expect. Like you said, we don't like tests that don't confirm our suspicions, but can you get a 'bad scan'?

We shot the RED 4k 2:1 and though this gave us a bit more horizontal resolution than the 35mm, the resolution chart should still be accurate because they fill the vertical frame, but not the horizontal on all the formats.

The tech specs were listed with the cml post, but I'll append a page to the paper.

Thanks for your critique David, we're big fans.

Benjamin Verhulst
benjaminverhulst.com

 

[David Mullen ASC]

Thanks for the info. In terms of the difference between the 35mm scan, the Red frame, and the Super-16, I would have to suggest that you talk it over with someone at FotoKem, perhaps even get them to put all three up in their D.I. theater screen so you can play around with them.

But I suppose a rough answer would be that the film frame was overexposed (perhaps more than 2/3's of a stop) but the RED frame was relatively underexposed in the sense that you protected the highlights to the degree that the result matched the overexposed film negative. But then that doesn't explain why the the film footage was lacking in shadow detail in the night stuff if you were being so generous with the exposure. On the other hand, these examples were all color-corrected so I can't really tell what the originals were like. It may be interesting to see the uncorrected LOG version of these images. To my eye, the 35mm footage was timed to look more contrasty than the RED footage, judging by the hotness of the shirt, etc.

Being old-school, the easiest way for me to know if the negative was too dense would be to look at the printer lights of a balanced answer print.

It's a good paper and I'm sure your faculty was impressed.

 

[M Most]

We expected to have to compensate for exposing the 5219 at 320 ISO, but there is really very little left in the highlights, which threw us in a spin. We triple checked and everything had been exposed properly. Again, the working print looked fine and had all the detail we'd expect. Like you said, we don't like tests that don't confirm our suspicions, but can you get a 'bad scan'?

You can. But nowhere in your paper does it indicate that you used a proper print preview conversion LUT when you color corrected the film material. From what I can see, you simply took it into Apple Color and changed the settings until it looked "right." But the fact is that you're never going to get an accurate image by doing that. If you're correcting from a Cineon file, you must use a conversion LUT to put the values at a proper starting point for electronic display and proper contrast. I would have thought Rex would have insisted on that...... :001_unsure:

 

[Zach Gray]

I think the most notable difference (aside from DR) in RED vs Film comparisons is frequently ignored.

It's the luminosity of the halftones. It's the 'glow' of skin or surface just before it turns away from the light.

Painters have long understood that the true color of an object is revealed in this zone between the core shadow and the highlight. In my opinion, it's what makes film (and painting) look alive.

I think RED does a pretty good job of being accurate.

But if you look carefully at the halftone and shadow saturation of film - it's amazing and rich.

Even in the youtube clips he posted.

http://www.youtube.com/watch?v=SNFI76bkhFQ

Not really perfect or fair comparisons. Just my thoughts after looking closely at a bunch of footage.

Is it grading? From all the RED footage I've seen and shot, this condition can happen, but it's hard to get. RED tends to gray and flatten in the halftone and desaturate in the shadow. It may be a more accurate representation of reality than film, but it's not as rich as what we are used to.

I know there are fundamental differences between silver and silicon, but if RED cracks this one in their sensor design...