4K, 8K and Lenses

[Christoffer Glans]

Coming out of the screening for the 70mm version of Hateful Eight I saw that the sharpness weren't really there. It looked like a 35mm screening because of the old lenses used. Yes, the grain was smaller and the field of view/depth of focus were really good looking, but the 12K equivalent resolution of 65mm film weren't there at all and it's because of the lenses. I wouldn't have changed the lenses because they created that specific look, but the question that came up was... why 4K if lenses do not cover it? People speak of lenses being too sharp, that they want some softness to them, but if so then they will not cover the resolution of 4K. When I bought my set of CN-Es I really loved that they cover 4K sharpness and you really get a true 4K resolution image, with all those minor details you would want 4K to show. But using other lenses the resulting resolution sharpness rarely even cover 3K.

So, while Hateful Eight were all about giving that old Panorama quality, shooting 4K, 6K or 8K and above, what is the purpose of that other than for downscale into nicer debayered samplings? If we are to do 4K deliveries, we need lenses that really give that sharpness otherwise, what's the point of that resolution? Wouldn't 2K or 3K shooting and later upscaling to 4K be rather closer to what most lenses cover anyway? If the lenses aren't sharp enough for 4K, there won't be any details of lens characteristics to resolve anyway.

So what lenses ACTUALLY covers 4K resolution sharpness? Which lenses with a desirable character still holds sharpness in higher resolution and so on? I know my Canon CN-Es does it, but can we do a proper list of this?

 

[Phil Holland]

The highest resolving lenses I've tested and shot with thus far in terms of overall performance are (in no particular order):

- Leica Summilux-C
- Arri/Zeiss Master Primes
- Arri/Zeiss Ultra Primes
- Zeiss Otus Primes
- Sigma Art Primes
- Illumina S35 Primes (newest series, but sets do indeed vary)
- Panavision Primos
- Leica Summicron-C

Some of the above exceed 200lp/mm wide open in the center, field, and edges/corners. I've used all of these on the Dragon sensor at this point as well and many, many other lenses. 200lp/mm in the center isn't too uncommon for most lenses, but once you leave the center frame that's where many lenses start showing their worth in terms of resolving power. Dipping into the 140lp/mm range is pretty average, edges and corners in the 70lp/mm range is pretty meh performance.

I evaluate lenses for Super 35mm and FF35/VistaVision formats mainly. 4K is arbitrary unless you are tying it to a format size and pixel pitch. 5K in the S35 area on the Dragon sensor and exceeding that format size from 6K through now 8K is what I reference at this point. It's interesting seeing some of these lenses on smaller pixels than Dragon, but that opens up ugly bits if you go too far. Especially if you have motion picture work in mind. That's at least the state of things now. Perhaps not in the future however.

There's lots of interesting copy variation and all that if you are going for lenses outside of the box. I have a 35mm f/3.5 Mamiya lens that performs better than some of the above, but it's slow and not exactly part of a cinema set.

The lens selection for Hateful Eight had nothing to do with resolving power, but more of reviving the 1.25X 65mm anamorphic format and creating that look throughout the film.

If you want to really, really squeeze the meat out of Dragon explore different methods of scaling and processing. It makes a load of difference if "sharpness" is your concern. Remember REDCODE RAW has no image sharpening applied to it.

As for desirable character, whelp, that's an interesting discussion. I like both very sharp, high resolving power, high acutance/micro-contrast lenses as well as funky things that aren't exactly anywhere near that.

 

[Christoffer Glans]

The highest resolving lenses I've tested and shot with thus far in terms of overall performance are (in no particular order):

- Leica Summilux-C
- Arri/Zeiss Master Primes
- Arri/Zeiss Ultra Primes
- Zeiss Otus Primes
- Sigma Art Primes
- Illumina S35 Primes (newest series, but sets do indeed vary)
- Panavision Primos
- Leica Summicron-C

Some of the above exceed 200lp/mm wide open in the center, field, and edges/corners. I've used all of these on the Dragon sensor at this point as well and many, many other lenses. 200lp/mm in the center isn't too uncommon for most lenses, but once you leave the center frame that's where many lenses start showing their worth in terms of resolving power. Dipping into the 140lp/mm range is pretty average, edges and corners in the 70lp/mm range is pretty meh performance.

I evaluate lenses for Super 35mm and FF35/VistaVision formats mainly. 4K is arbitrary unless you are tying it to a format size and pixel pitch. 5K in the S35 area on the Dragon sensor and exceeding that format size from 6K through now 8K is what I reference at this point. It's interesting seeing some of these lenses on smaller pixels than Dragon, but that opens up ugly bits if you go too far. Especially if you have motion picture work in mind. That's at least the state of things now. Perhaps not in the future however.

There's lots of interesting copy variation and all that if you are going for lenses outside of the box. I have a 35mm f/3.5 Mamiya lens that performs better than some of the above, but it's slow and not exactly part of a cinema set.

The lens selection for Hateful Eight had nothing to do with resolving power, but more of reviving the 1.25X 65mm anamorphic format and creating that look throughout the film.

If you want to really, really squeeze the meat out of Dragon explore different methods of scaling and processing. It makes a load of difference if "sharpness" is your concern. Remember REDCODE RAW has no image sharpening applied to it.

As for desirable character, whelp, that's an interesting discussion. I like both very sharp, high resolving power, high acutance/micro-contrast lenses as well as funky things that aren't exactly anywhere near that.

Me too, I like lens character when I see it, but I also love sharpness when I see it. I know that the Hateful Eight didn't have sharpness in mind but the look and feel of the old 65mm movies, so that wasn't the problem, it wasn't really a problem at all, just something that came to mind when watching it; "huh, the resolution and sharpness looks like usual 35mm projection" was my initial thought. Of course there was a difference but not that much.

As for my CN-Es they resolve sharp resolution much better than most lenses I've tried. I've only tried the masters and leicas a few times so I never really looked hard on the sharpness part for them. As for my Sigma ART it's not really matching the resolution of my CN-Es I'm afraid.

As for the Redcode, I've worked with it for 8 years now with various lenses, so I know the processing bit, but this has more to do with lenses resolving quality. Mostly I'm thinking about the fact that many use Dragons with older lenses and in my opinion I'm not sure what's the point of that. Better then to do that with Arri cameras of lower resolution, cause they will not resolve any greater anyway. Those lenses have a look, and that look is what's desired, just like with Hateful Eights Panavision set. But if 4K is the delivery point and you shoot 6K or 8K to resolve down to 4K for ultimate sharpness result, then lenses with a more "perfect" resolving quality is needed. I guess that medium format lenses as the Mamiya lens you mentioned would resolve best of all really. But as for cinema lenses, I'm thinking the Summicron is closest when shooting 6K? I'm not sure about the Master and Ultras really, they are good looking, but do they come close to the Leicas? The Otus is a monster though.

 

[Peter Lyons Collister ASC]

In the 70's Kodak kept coming out with sharper and faster emulsions. With Panavision and Zeiss coming out with sharper lenses at the same time. Our dear recently departed DP Vilmos started pre and post fogging film, using filters and smoke to degrade the image. Look at "McCabe and Mrs. Miller and it is an amazing piece of cinematography. So he was not using the full resolving power of 35mm film. But he did not consider shooting on 16mm. Same as Bob Richardson shooting 65mm but using those rare PV lenses which are not sharp and certainly not able to resolve what 65mm emulsion can give but gave the film a quality that he and Quentin were after, with the circle of confusion DOF falloff.
Trading sharpness and resolution for texture and story telling is a fair tradeoff

 

[Ignacio Aguilar]

Same as Bob Richardson shooting 65mm but using those rare PV lenses which are not sharp and certainly not able to resolve what 65mm emulsion can give but gave the film a quality that he and Quentin were after, with the circle of confusion DOF fallofff

Yes, if Tarantino and Richardson had wanted an ultra-sharp 65mm film, they probably would have chosen 5-perf 65mm spherical -maybe with the new Panavision Primo 70 lenses-, but they liked the particular look of the old MGM Camera 65 / Ultra Panavision lenses, even these aren't as sharp as spherical optics. I've seen 5-perf 65mm projected and DCPs from several 5-perf 65mm films, and even at 2K resolution "West Side Story" (65mm spherical) for instance looked sharper than "Ben-Hur" (65mm anamorphic).

 

[Jacek Zakowicz]

Any decent modern still lens resolves equivalent of 8K today. And that's for still photography where flat field is scrutinized for hours as opposed to motion when it can only be seen for 1/24th of a second haha ;-) that's your lens answer. Everything else is the good old fashioned industry snob-ism... not that it can be ignored- it's moved mountains so to speak....

 

[Christopher Probst]

Dragon 6K tops out at 100 lp/mm resloving power and based on the photosite ("pixel") pitch of that sensor... And prehaps surprisign to some of you out there, the same holds true for Weapon 8K. That resolving power is just spread out over a larger imaging area.

I'll explain... The easiest and most basic way to think about it is: You need 2 photosites to capture a line-pair off a chart. One photosite to get a black part of a test pattern, one photosite to capture the white next to it. So 6K is 6144 photosites across (not taking into account Bayer pattern or anthing on that level). So to capture a line pair you need 2 photosites, that means 6144/2 = 3072 pairs. Again, resolving power is measured in line-pairs per milimeter... And at 6K, the Dragon sensor is 30.7mm across... So 3072 line-pairs across a 30.7mm sensors is 3072/30.7 = 100.065 lp/mm...

Do that math with 8K and it's exactly the same resolving power. 8192/2 = 4096. 4096/40.96mm = 100 lp/mm. So until the photosite density is increased with a new sensor, Red Dragon sensors are capable of resolving 100 line pairs any day of the week. Now having a lens like a Master Prime or a Summilux that out resolves the sensor is a good thing when you take Nyquist sampling therom into account...

 

[Michael Tiemann]

Dragon 6K tops out at 100 lp/mm resloving power and based on the photosite ("pixel") pitch of that sensor... And prehaps surprisign to some of you out there, the same holds true for Weapon 8K. That resolving power is just spread out over a larger imaging area.

I'll explain... The easiest and most basic way to think about it is: You need 2 photosites to capture a line-pair off a chart. One photosite to get a black part of a test pattern, one photosite to capture the white next to it. So 6K is 6144 photosites across (not taking into account Bayer pattern or anthing on that level). So to capture a line pair you need 2 photosites, that means 6144/2 = 3072 pairs. Again, resolving power is measured in line-pairs per milimeter... And at 6K, the Dragon sensor is 30.7mm across... So 3072 line-pairs across a 30.7mm sensors is 3072/30.7 = 100.065 lp/mm...

Do that math with 8K and it's exactly the same resolving power. 8192/2 = 4096. 4096/40.96mm = 100 lp/mm. So until the photosite density is increased with a new sensor, Red Dragon sensors are capable of resolving 100 line pairs any day of the week. Now having a lens like a Master Prime or a Summilux that out resolves the sensor is a good thing when you take Nyquist sampling therom into account...

Christopher,

In another thread this question Graeme Nattress weighed in with an estimate of 100 px/mm, not 100 lp/mm. If we agree to ignore the Bayer pattern (which Graeme argues can reach approx 70% of monochrome) the fact is that if every line signal is perfectly aligned with every pixel, then you are correct that each pixel pair can perfectly resolve the line. However, if we are unlucky and every line falls precisely between two pixels, then we perfectly resolve 50% gray (i.e., nothing at all). The posting that started the other thread looks at the question of how many actual pixels are needed so that a line pair is properly resolved (to a given MTF value) regardless of whether it lands perfectly across the pixels or whether it is shifted by 1/2 a pixel in one direction or another.

In the real world, lenses don't generally resolve to MTF100. In many cases we are happy when they remain above MTF50. When a Master Prime claims 200 lp/mm and some other lens only manages 100 lp/mm, neither lens is actually reporting how many lp/mm it can resolve with a 100% MTF. But in general, if two lenses publish their best measurable lp/mm at *some* MTF, and then they are compared at a constant lp/mm rating, such as 100 lp/mm, we should expect the one with the higher lp/mm resolution to have better contrast than the one with the lower lp/mm. Which goes right to your point that the greater the resolution of the lens, the better the resulting image will be, even with a dramatically lower resolution rating for the sensor.

In the final analysis, any lens that has a good MTF rating at 100 lp/mm is going to be a great lens for DRAGON. But a lens than can resolve 100 lp/mm only at a very low MTF rating might disappoint.

 

[Marc Wielage]

Trading sharpness and resolution for texture and story telling is a fair tradeoff

I think this is very true. And in post, we often wind up up softening images -- particularly skintone -- far more than we ever sharpen anything. Sharpness is not going to tell the story any better if it winds up making the actors look more haggard and less attractive.

 

[Nick Morrison]

Dragon 6K tops out at 100 lp/mm resloving power and based on the photosite ("pixel") pitch of that sensor... And prehaps surprisign to some of you out there, the same holds true for Weapon 8K. That resolving power is just spread out over a larger imaging area.

I'll explain... The easiest and most basic way to think about it is: You need 2 photosites to capture a line-pair off a chart. One photosite to get a black part of a test pattern, one photosite to capture the white next to it. So 6K is 6144 photosites across (not taking into account Bayer pattern or anthing on that level). So to capture a line pair you need 2 photosites, that means 6144/2 = 3072 pairs. Again, resolving power is measured in line-pairs per milimeter... And at 6K, the Dragon sensor is 30.7mm across... So 3072 line-pairs across a 30.7mm sensors is 3072/30.7 = 100.065 lp/mm...

Do that math with 8K and it's exactly the same resolving power. 8192/2 = 4096. 4096/40.96mm = 100 lp/mm. So until the photosite density is increased with a new sensor, Red Dragon sensors are capable of resolving 100 line pairs any day of the week. Now having a lens like a Master Prime or a Summilux that out resolves the sensor is a good thing when you take Nyquist sampling therom into account...

Love this post. Thanks for sharing Christopher.

 

[Michael Tiemann]

Here is another thread that looks at resolution vs. pixels, started by Jim Jannard back in 2010.

There is some confusion because the graph labels (4K (?)) and image descriptions (2K (?)) seem to disagree, but if we abstract it all out, what we see is that when there are lots and lots of pixels available per line element (in this case, sinusoidal concentric circles), we see that there is indeed a drop in MTF as the width of the rings thins out. The Alexa gives up at the 1500 (750 pixel? line pair?) mark, whereas EPIC MX hangs on strong through the 2500 (1250 pixel? line pair?) mark. The Alexa goes more or less to mush after about 1750, whereas the RED drops, but doesn't disappear from 2500-3500, and still shows what looks like a detectable signal out to 4000 (albeit at an MTF of around 10). I think there is no doubt that RED offers a high-resolution platform, but the relatively finite number of actual rings that can be counted with a relatively high contrast ratios suggests that even a high resolution platform is not pixel-perfect. But RED's real-world performance is a thing of beauty.

 

[PatrickFaith]

I think this is very true. And in post, we often wind up up softening images -- particularly skintone -- far more than we ever sharpen anything. Sharpness is not going to tell the story any better if it winds up making the actors look more haggard and less attractive.

I agree and have to soften a lot of things so it doesn't pull the audiences eye to a non-story component (i.e. female hair on thier neck), on the other hand sometimes sharpness does tell a story but I've noticed it's very small sections of the frame that needs to be uber sharp. Would be nice in 8k for colorist during secondary grading to be able to "sharpen" an area with the full 8k wavelet, while otherwise just using a softer wavelet transform.

 

[Jacek Zakowicz]

I agree and have to soften a lot of things so it doesn't pull the audiences eye to a non-story component (i.e. female hair on thier neck), on the other hand sometimes sharpness does tell a story but I've noticed it's very small sections of the frame that needs to be uber sharp. Would be nice in 8k for colorist during secondary grading to be able to "sharpen" an area with the full 8k wavelet, while otherwise just using a softer wavelet transform.

HAhahaha I want to see that !!!!
They can't even figure out how to put flares or anamorphic look in post so the production has to use shitty glass, uncoated glass, anamorphic glass, etc and you expect them to spot sharpen the image !!!??