Steve Yedlin Resolution Demo

Hello, I don't know if any of you follow Steve Yedlin, ASC. on Twitter, but he recently shared this video demo on various resolutions of various cameras. Who is Steve Yedlin you ask? He's the cinematographer behind Star Wars Episode VIII: The Last Jedi, Looper, The Brothers Bloom, and Brick. You can also follow his Twitter here: https://twitter.com/steveyedlin

Tweet

Screen Shot 2017-07-24 at 3.19.23 PM.png (920.42 KiB) Viewed 20887 times

As he wrote on his page regarding the demo: "This demo was designed to be viewed in full raster with visually lossless image files, but is presented here in compressed format to make possible wider distribution."

"To find out about theatrical screenings in the intended format or for other inquires: displayprepdemo@yedlin.net"

The demo is in two parts. You can see the demo here: http://yedlin.net/ResDemo/

I think it's worth checking out because he also says in the voice over some very interesting points. I tend to agree.

The goal of the camera manufacturers should not be about more pixels, but smarter and better pixels. The goal should be more dynamic range, better color science, and improved shutter such as global shutter. This is often why I say I hope Blackmagic gets the 15-stops of dynamic range in the 4.6K to work in a global shutter and to focus less on the higher resolution push.

Either way, check out the videos. Leave comments and thoughts below.

I have to agree more megapixels is not the answer. In y own testing I have found the 4.6k to resolve nearly as much as the 8k helium. It's really quite negligible. We were very surprised. JB in another thread discussed how arrow still has the same sensor but continued to refine it over the years to squeeze every last drop out of it. I hope it's the direction black magic moves with the 4.6k as it's quite a fantastic sensor

Thanks for sharing this!

Very interesting indeed! Thanks for sharing.

No problem guys. I just found the American Cinematographer write up interview on the whole demo: https://theasc.com/articles/a-clear-loo ... resolution

And further reading from Yedlin on color science: http://www.yedlin.net/OnColorScience/

First off that was a fun watch, Thanks for posting. But, it almost sounds like he's talking to the type of people who never realize you can have all three components in a image, Great DR, Awesome colors, Annnnd High Resolution. They aren't some how mutually exclusive to each other.

(Just a fun question: Why did he say he wanted to put film emulsion grain on all test images to even the results out?) Film grain is a flaw in film, nostalgic, but still a flaw that hides some of the resolution. Why cripple the non film grain mediums?

I'm glad he emphasizes several times that he's talking about cinema projection. I haven't seen a projector that can match Color, HDR, or Resolution of a average OLED 4K TV. I literally go to the theater twice a year mostly for new Star Wars episodes. VFX that looked very good in theaters look several degrees not as good on my TV . My home experience is just so much better and cheaper.

Theaters are dying! Projection is dying! He maybe right when saying 4K is retina, especially on a large projected screen (Funny how he doesnt even test 1080 lol. Almost like he saying, 'get that trash out of here'.
But, I can easily tell the difference of 1080x1920 (2K) to 4K on my TV it's easy I don't even have to look more than a sec. Or even 1920 (2k) upscaled to 4K it's pretty noticeable. I can't comment and say 8K is any better since I don't have a native 8K monitor or TV.

4k maybe the magic resolution setting! But I can tell you that 1080 (2K) is not it, at least when viewing at home at a normal couch to TV distance, and on a Simi decent 4K TV.
Theater projection? He sounds right on the money, but theater projection can't give me the same DR, colors, or resolution my 4K OLED can so.



VVVVVVVV. Below is just me ranting. VVVVVVVV

Honestly I'm glad theaters are dying, high prices, people aren't courteous anymore, high prices. Easier for Indies, true Indies to get a film seen by everyone. Studios don't determine what is "Pop" or what they think I should watch. (As much as before) Netflix still has that power. Radio stations use to have that control too.
Pay for View on Vimeo doesn't have that power at all, You want to upload to the "purchase" area of Vimeo and charge, feel free. NO camera stipulations needed. And I'll tell you there are some great hidden gems of movies on Vimeo PFV area.

Explanation about Halation of Film emulsion was really new thing to me. I always noticed that something occurs around overexposed highlights in film but never can't catch the nature of it.
Worth to watch for deeper understanding what film look consists of!

Very interesting demo. I enjoyed it a lot. To me this whole conversation in reminiscent of the early days of the digital SLR. I started off photographing in 2002 with a 6mp Canon D60 and 4mp 1d. The whole discussion back then was that there wasn't much need for higher resolution, how well the images upsampled, digital could never match scanned film etc. etc. The digital cinema camera industry has the same megapixel race but moves at a much, much slower pace. Basically, these tests and conversations don't amount to much. Even if it's true, the industry will push higher and higher resolutions cameras. Prices will drop and higher resolutions will become the norm. 10 years from now we'll all probably be shooting 8k+ whether or not it's necessary!

Well worth the time to watch Steve's testing and explanations. You will probably learn a great deal. I certainly did. Very heartening to hear that from a pixel resolution count we're probably already at a quality point necessary with 2k to 4k resolution and other factors are more influential. No need to continue the resolution chase madness and storage/bandwidth upgrades and playback issues.

Disheartening was the continuing end results from typical delivery playback of compression and true delivery from satellite. cable, internet providers such as Netflix/Amazon.

Again, an interesting scientific practical look at the current state of the resolution issues.

On a scale of 1 to 10, ranking the ten most important criteria contributing to the success of a cinematic production, I suspect acquisition resolution gets a zero. If it's a hair commercial or architecture documentary, 1. Top of the list: emotion.


Sent from my iPhone using Tapatalk

Interesting.

On a scale of 1 to 10, ranking the ten most important criteria contributing to the success of a cinematic production, I suspect acquisition resolution gets a zero.

Here's a still from the most profitable feature film of all time:

On a scale of 1 to 10, ranking the ten most important criteria contributing to the success of a cinematic production, I suspect acquisition resolution gets a zero. If it's a hair commercial or architecture documentary, 1. Top of the list: emotion.


Sent from my iPhone using Tapatalk

Agreed Rick. All you have to do is have the camera at the right place at the right time. It's as simple as buy low and sell high.

First off that was a fun watch, Thanks for posting. But, it almost sounds like he's talking to the type of people who never realize you can have all three components in a image, Great DR, Awesome colors, Annnnd High Resolution. They aren't some how mutually exclusive to each other.

He never says they are mutually exclusive. He does point out there there is a direct tradeoff between increasing noise and increasing pixel density. At any particular sensor size, you can't simply increase the K count without also degrading the image through increasing noise.

Film grain is a flaw in film, nostalgic, but still a flaw that hides some of the resolution. Why cripple the non film grain mediums?

Yedlin clearly acknowledged that grain degrades the image. He also clearly explained that in his experience many viewers will report that an image with added grain is perceptually sharper than the same image without grain. He offers his own theories as to why this may be true. All those pixels on the screen, degraded or not, are filtered through our eyes and brains, which can yield results contrary to the actual facts of the image. This is why he is constantly careful to state when he is talking about "perceptual sharpness." So, yes, grain is a degradation, but our brains don't always see it that way. Take a look at any optical illusion if you need proof

First off that was a fun watch, Thanks for posting. But, it almost sounds like he's talking to the type of people who never realize you can have all three components in a image, Great DR, Awesome colors, Annnnd High Resolution. They aren't some how mutually exclusive to each other.

He never says they are mutually exclusive. He does point out there there is a direct tradeoff between increasing noise and increasing pixel density. At any particular sensor size, you can't simply increase the K count without also degrading the image through increasing noise.

Film grain is a flaw in film, nostalgic, but still a flaw that hides some of the resolution. Why cripple the non film grain mediums?

Yedlin clearly acknowledged that grain degrades the image. He also clearly explained that in his experience many viewers will report that an image with added grain is perceptually sharper than the same image without grain. He offers his own theories as to why this may be true. All those pixels on the screen, degraded or not, are filtered through our eyes and brains, which can yield results contrary to the actual facts of the image. This is why he is constantly careful to state when he is talking about "perceptual sharpness." So, yes, grain is a degradation, but our brains don't always see it that way. Take a look at any optical illusion if you need proof

I'm with this guy mostly, I just don't agree with everything he is saying because I have two eyes and have noticed differences that he say are not noticeable. (Lucky me right?)
I know this guy has a lot more knowledge than me on the subject and his main goal was to say, hey we have enough K's let worry about compression and the such now. And I even said maybe 4K is the magic number.
And I've always thought compression is the main killer of a great image. Look at some of my past BMCuser comments.
I just feel that intro felt a little more like he thinks his audience doesn't know DR, Color, Compression, etc. trumps Resolution. I've never blindly followed wha anyone has said no matter how experienced they are, but their experience level does weigh more heavy on my thought process and understanding in the long run.
I've done resolution test on RAW personal footage, forward and back, back and forth and I tell you I can see differences that he say nobody will ever see. No insult to Mr. Yedlin. But none of that footage was on a big screen so I maybe missing the point. :0


I'm not completely believing your comment on film grain adds to perceived resolution, really I'm not buying it at all. I'm one of those viewers he is talking about, and I've never heard anyone say that.

And just because there are optical illusions out there that trick our brains doesn't support his claim or mean it's valid in any aspect.

I literally was watching something this afternoon that made me notice just the opposite.
The original, (Swedish version?) "Girl With The Dragon Tattoo" series, in the third film "The Girl That Kicks The Hornets Nest" throughout the film they use a lot of film grain, either with a heavy grain film stock or added in post digitally.
There is one scene where Lisbeth is in court fighting for her freedom, and I'm not sure if they used two separate cameras or the colorist forgot to add grain to those scenes, but while showing Lisbeth the grain is heavy and to the point of distracting. But when it cuts to her psychiatrist and prosecutor, across the room, the grain is gone and the image looks super sharp and crisp. These scenes are cutting back and forth with each other one after another so it's even more noticeable. Funny that I literally watched the third movie in the trilogy today.
I can tell you the non grain scene look incredibly sharper a clearer.
Or maybe they had a better camera or film stock, either way less grain look a lot better.

So yes I'm calling BS on the perceived resolution added by film grain. It's on Netflix right now so anyone can check it out and notice it.
Or hey my brain isn't getting trick by optical illusions.


One more thing why do Sony, Red, and even BMD sensors get cleaner in the noise area when they add more photosites to the same relative sensor sizes? Seriously asking at this point.

All in all though everything he posted and explained was well worth the time, and I do appreciate you clearing some things up.

In defense of resolution, it's one of the constants in describing a CMOS camera.

Excepting some rare examples where sensor area is locked by a license, resolution is fixed to a camera. Dynamic range and noise change with settings, but resolution is constant, so it's the line in the spec sheet that is more reliable. I understand why the marketing is set up the way it is.

I thought his video was excellent. The example frames of grain vs no grain clearly show that perceived sharpness increases with some grain. I'm sure there's an upper limit but his example was really cool.

His point is that many things contribute to perceived sharpness and over 2k resolution, these other factors are more important than increasing the resolution. I think he demonstrates it quite convincingly

Yedlin is working in Nuke to do his scaling transformations and he stressed that the quality of those transformations is crucial. Is anyone familiar with Nuke to know which he was using, and whether the transforms in Resolve, especially the upscale, are of equal quality?
I've combed through the Resolve manual, but nowhere does it explain what algorithms aren being used when you set the scaling options to "smoother" or "sharper" in the preferences. Presumably it's some secret sauce, but I'd prefer it to be transparent to the user. I'm guessing Yedlin is using Nuke because you can select the exact algorithm used for scaling by name:

http://help.thefoundry.co.uk/nuke/8.0/content/user_guide/transforming_elements/filtering_algorithm_2d.html

In the part two video you can see he is using the Simon algorithm- I was interested myself so I kept my eyes peeled

But that's as far as My knowledge goes on this!

...
Yedlin clearly acknowledged that grain degrades the image. He also clearly explained that in his experience many viewers will report that an image with added grain is perceptually sharper than the same image without grain...

That is very interesting. It is well known that audio noise can have the effect of making faint sounds more audible. I wonder if the same mechanism is at play in the visual cortex given the similar neuronal structures throughout the brain.

One thing Mr. Yedlin's demo has made me really understand from multiple past experiences, and almost confirm, is that there are so many levels of a human's vision resolving power. And that we really are finding out that some, like all other human functions, genetically posses different levels of natural human biometrics.

Please bear with me on this short foreshadow to what I'm talking about, and conclusion I'm starting to reach through my own personal experiences now later in my life. 40's
As a kid you won't notice it and you don't really have the conscience level to start thinking on that level in the first place. But, in my early adulthood as a soldier in the Infantry, I would notice while out on night patrols, in dense wooded areas or jungle that I was able to resolve detail at night a lot better than my squad mates, minus one other.
For example(One of many):On a moonless night, I remember halting the squad at the edge of the tree line over looking a large field maybe 300 meters across. 100 to 150 meters into the field there was a single tree. I let the other guys know there was someone standing under the tree. The only reply I got from everyone was, 'What tree?' lol. Only one other guy said, yeah I see him. We finally all put on our PVS7s and it became obvious for everyone.
I also had instances of color variance getting me into friendly arguments.
A cycling rep had given me a pair of his company's new cycling kicks to try out and wear for the season. I remember giving him a hard time because one was a different shade of red than the other. Everyone in the shop was acting like I was crazy, but I eventually dropped it thinking they saw it but were just messing with me. lol

From time to time the "Cinema Projection" arguments come up on different forums and you'll get one or two guys who take a very aggressive stance that they can easily see the difference between 2K projection and 4K projection, almost like it's night a day difference. You also get forum members across the cinema world who say that they don't understand why others don't see the COLOR differences in certain images. In most of those instances they just get the cliche "whatever" reply from everyone else. lol

So my conclusion is just like every other form of evolution, running sub 10sec 100 meters, high jumping over 8ft, holding fifty moves in your head from a chess tournament, stopping a 300lb line man... etc. etc.
Well it looks like some of us just have more 'Rods' and "Cones' in our Retinas and we can resolve more detail and see more shades of color variations.

If the image from a few years ago, of the Blue&Black/White&Gold dress has thought us anything, is that we can all be looking at the exact same object or image and see something completely different. lol By the way I see the dress as white and gold. Some on her are actually seeing this exact picture as a blue and black dress. weird

IMG_1251.JPG (53.33 KiB) Viewed 20289 times

So Mr Yedlin is right on all accounts, and the people who are saying they clearly see difference are right too. You never know what someone is really seeing, and have to take them for their word. If just ten percent of the almost 8 Billion people on the plant have I higher levels of rods and cones that's almost 800 million people world wide. and if one percent of those have a even higher elevated level of those same rods and cones, well you get my drift.

I guess what I'm saying is that I always have thought cinema looks like poo compared to 4K TVs. And film grain has always been a giant distraction to me and really makes films that over do it almost unwatchable.
And most of my past comment either here or on BMCuser have implied the same thing. And I know I'm not the only one either, There are constantly members on all the other forums who always act surprised when most people say they can't tell the difference or just barely notice. To us it looks very noticeable.

All these resolution test done across the internet are very interesting because you always get a pretty wide spectrum of comments on people perceived views on it.

So hey you might as well future proof your footage for people who will be viewing it a thousand years from now. Just kidding on that last one I had to throw that in though.

Add On: There are cases of people with more sensitive hearing as well and usually aren't believe until they are put in sound booths and frequencies that are supposedly inaudible for humans are played through head phones, like in a hearing test, and are detected by these individuals. Some of these individuals I hear (no pun) are in the music industry.

The main issue with all these resolution test and demos is that we are assuming everyone on the planet visually resolve detail and colors on the same level as everyone else. And we don't not even close.

In the part two video you can see he is using the Simon algorithm- I was interested myself so I kept my eyes peeled

But that's as far as My knowledge goes on this!

I noticed that as well, but I'm not familiar enough with Nuke to know which node that it is reporting. I didn't notice it changing dynamically as he turned nodes on and off. So it could be reporting the setting in nodes he was using as examples of poor transforms for all I know.

If the image from a few years ago, of the Blue&Black/White&Gold dress has thought us anything, is that we can all be looking at the exact same object or image and see something completely different. lol By the way I see the dress as white and gold. Some on her are actually seeing this exact picture as a blue and black dress. weird

IMG_1251.JPG

I still bring that damn dress image into Photoshop and use the color selector and get Blue & Black. I've taken screenshots and posted to Facebook to prove to people that corrected the image shows a Blue & Black dress, and that uncorrected the color selector still shows those as the colors of the dress. That stupid dress was the most obnoxious thing because it was a terrible photo to begin with and needed to be properly balanced.

I digress from going back into the dark hole of debating why that dress photo is stupid as ever since the image was so poorly photographed to begin with. Your points are valid. People perceive the world differently.

Early on I questioned the jump to 4K, but after seeing the difference of a HD TV next to a UHD TV I was convinced that UHD 4K was better. However, since it is plenty for the home I say we skip 8K pushes for televisions and focus on the HDR standard within the 4K televisions. Also, we shouldn't be pushing 8K UHDTV when no broadcaster supports 4K UHD yet. The only places you can watch 4K UHD content are Netflix, Amazon Prime, and... oh wait that's about it because then it's YouTube & Vimeo.

I personally don't mind higher resolution acquisition formats for a 4K DCP or 4K UHD output. It's important for image stabilization, reframing, and VFX work. However, there's a limit. Until we have a digital acquisition format that matches the amazing quality of 15-perf 70mm IMAX film and a digital projection technology that matches that quality. Now, in the test Yedlin states the 15-perf 70mm film was scanned at 12K. I think I've read somewhere before that 15-perf IMAX film is equivalent to 16K resolution. So...

Once we can digitally project 16K on a big 1.43:1 IMAX screen then we can have a camera manufacturer push that resolution. However, the sensor size should equal the film gauge of 15-perf IMAX film. So that then becomes the next point to talk about... sensor size vs resolution. Now that large format and large resolution format should be reserved for the largest screen possible, which is the movie theater. I'm someone who personally still loves the movie theater. So I'd love to see it continue to thrive.

his main point though was that its not the ultimate theoretical resolution given by the format but the whole image acquisition system that matters

in this case the 6k from the Alexa 65 had much more perceptual resolution than the 11k imax film scan - I don't think it being scanned higher would have led to any more

A lot of his discussion was really about the fact that most real world acquisition systems don't come close to maxing out the potential resolution of the format they are captured at

He also makes the important point that the size of whatever screen we're watching probably won't change much anymore. I.e. Most TVs are X inches diagonal and Y feet from the couch, therefore projecting an image of a certain size on our retinas. If we're already at the point where (most people's) eyes can't resolve individual pixels at that distance, then subdividing those already too small to see pixels even more is useless.

I never go to theaters so don't have much to say about screens that size.

TV manufacturers should focus on dynamic range in my opinion

Re: Steve Yedlin Resolution Demo
by Jamie LeJeune » Thu Jul 27, 2017 4:01 am

Gavin_c_clark wrote:
In the part two video you can see he is using the Simon algorithm- I was interested myself so I kept my eyes peeled

But that's as far as My knowledge goes on this!


I noticed that as well, but I'm not familiar enough with Nuke to know which node that it is reporting. I didn't notice it changing dynamically as he turned nodes on and off. So it could be reporting the setting in nodes he was using as examples of poor transforms for all I know.

I emailed Yedlin to ask. This was his response:

"""Which is the “better” one? They’re just different, I don’t know about better/worse.

Resizing algorithms range from smooth to sharp. In the 2k/6k section of Part 2 of the demo, I’m comparing a smoother one to a sharper one. “Better” is opinion — I personally think the smoother one looks better, but the sharper one is the one that would trick more people that it’s “resolute” (the perceptual attributes associated with the work “resolute”, not actual resolution). But I personally think that one actually looks kind of artificially affected.

As far as I know, Nuke’s algorithms are not proprietary to Nuke and you can even look up the underlying math on the internet if you don’t have Nuke — I know most of them are public even if some might be proprietary because I know developers who have used the same algorithms in other software. I would guess that Resolve’s two options are just two of the ten options available in Nuke (even though they don’t tell you which they are).

As far as I know, the only ones that might be crappy (i.e. fall into a better/worse binary as opposed to just different) would be consumer resizing that happens on the fly. Like if your DVD player up-rezes to HD in real time while you play a disk. The only other thing I know of that would be bad as opposed to different is if you accidentally use “nearest pixel” (which in Nuke is called “Impulse”) — “nearest pixel” is the sort of a do-nothing algorithm. It just selects an already-existing nearest pixel value instead of interpolating the results. Something like that should only be used in the viewer or for a temp render (if you’re doing a temp and just need a faster render), not for proper mastering.

I’m attaching graphs of the algorithms. You can see that the sharp ones like Rifman and Simon have little inversions (where the plot goes down before it goes up), which is what makes the edges look sharp and contrasty. The smooth ones (like Notch and Parzen) don’t have that. Generally speaking, I like to use one of the sharper ones (but not the absolute sharpest) for NORMAL resizing (like 4k—>2k, 2k—>4k, 4k—>4k) but a smooth one like Notch for HUGE up-rezing (like doing a 20x magnification or taking standard-def and blowing it up to 4k)."""

I don't know how to attach images here

I emailed Yedlin to ask.

Thank you Francis! That's very helpful info

I was working on a VFX shot today and couldn't work out why a repainted background was 'sharper' than the original surrounds, turned out I had accidently bumped the intensity of the grain in the shot, set it back to where it was supposed to be and it was all good.

Steve Yedlin added responses to questions about his test:

http://www.yedlin.net/170728.html

How do you implement a scaling algorithm in Davinci Reslove?

How do you implement a scaling algorithm in Davinci Reslove?

you don't - you use the ones that are built in

same with Nuke really except it has more options there and I guess its at least theoretically possible to roll your own in that environment. For resolve I'd imagine you'd have to build a plugin of some kind