Life after Red

[shashbugu]

Thanks Stephen. Points taken except that I will not only listen to Graeme, but also to the Sony, Canon, Panasonic, Panavision, Arri and other companies I have had intensive talks with about sensor design, application and back end processing (De-bayering, etc). I am not one to take the word of a single camera manufacturer or its marketing / technical team. There are many ways to improve.

Cheers

Michael I am very disappointed in you.
I have read through all the posts on this thread, not a single person has attacked you personally. We have all passionately commented on the inaccuracies and the poor choice of post heading on your blog. I certainly don't believe that any poster on Reduser.net has the time of day to PM hate mail to you, especially over a recurring f65 vs Red Topic or Red vs everyone else. Its a daily thread starter nothing new here. We are desensitized to those topics and threads.
But we will always call you out if you are wrong or assumptive over issues and fud about the Epic we feel knowledgeable enough to rebuke.

I also highly doubt that you have personally discussed sensor debayering types with Sony, Arri, or Canon. You might have read it on one of their white papers, watched a seminar, or spoken to a few salesmen types at NAB or the other trade shows. I have looked at your blog, you don't seem to have any privileged insider information on most of the topics. I am not calling you a liar, that would be wrong of me and if it appears so I apologize. I often think many blogs these days are made up of regurgitated information.
Again if anyone has sent you hate PM's or mail, they are morons and should be treated as such. Please complain to the Mods...

Feature set check boxes don't lie.

[patrickortman]

Well, to me that's the point of the article: RED created something really cool. A lot of us embraced RED, and did well by our trust in them. Life after RED is much better! And now, thanks to RED, there's competition coming up. Choice is good. Competition is good, as long as it's fair.

He didn't at all go into the specifics of some peoples' problems with RED, although he did say RED needed a better attitude. RED has been gracious to me on a couple of occasions when I needed them to be. Certainly there's been a few well-publicized PR black eyes RED has gotten (Bloom?). But that's what you get when you're dealing with passionate people who love what they do, right? While SONY may not ban you from their forum for criticizing their products, with RED one can get the attention of Jim and Jarred and sometimes even effect change. It's a trade-off.

And some of his points about RED being a pain to work with on set and in post are absolutely true- shooting 4K is often totally overkill. That goes 4x for 5K. That's why I'm also buying a Blackmagic Cinema Camera or two, before I even think about replacing my still-working, still-makes-great pictures, utterly-reliable RED MX that even has some features the Scarlet is missing. And yes, I'm bummed that RED killed the original Scarlet, but I understand Jim's reasoning for doing so.

For me, due to RED's long delays in getting Epic Stage 3 happening, I decided (on Jarred's advice) to go ahead and upgrade my RED to MX with SSD and get started on my film and specs. It hurts a little, I'd really have liked to have had my Epic. But truth is, Epic wasn't ready, and by using the tools I had, and putting my Stage 3 money towards upgrades and a steadicam, I turned out a killer film and several commercials that are getting me a ton of work. I'm telling you guys this because you as a filmmaker need to use the right tool for the right job. Sometimes a C300 is a better choice than an Epic. Sometimes Epic is a better choice than Alexa. But the worst thing we can do is spend all our time obsessing on gear and forgetting that our jobs are to put out films. Sure, we're passionate about our cameras- for many of us, they're our most prized possessions. But putting RED on a pedestal and becoming immune to differing points of view is like those arch conservatives who only get their news on FOX (insert your own liberal example here). It's myopic. Cult-like.

And that's why I welcome what the other guys are doing with their cameras. And I expect that RED will continue to innovate, and I'll continue using their products when I'm working on a project where they're a great fit for me. Which is often.

[Michael N. Sutton]

Hi Graeme and Chris. I had made the revisions to my article but I think your last posts are based on what was written prior. An OPL filter (which the Epic has, as does most cameras) does trade off sharpness. There is no arguing that. Its called a blur filter by many for a reason. That is all I was saying, and have corrected it.

Graeme are you claiming that you DO NOT agree that Bayer pattern arrangement has an effect on image quality after interpolation? You keep mentioned the OPL glitch but not that for some reason. Sony and others seem to think it is important.

Detracting to a prior (and corrected) OPL tech reference that was less than 2% of the article doesn't change the overall point that everyone needs to step up their game which was the entire point.

[Michael N. Sutton]

Kudo's to Patrick for cutting to the point of the article. Competition is good. Red did something cool, others are following suit, Red now needs to counter, then they will need too, etc etc etc. Thank you.

[Martin Weiss]

If anyone has sent you hateful PM, please do forward it to one of us moderators, and we'll quickly and resolutely deal with the offender. Discussion is good, personal attacks are not.

[Martin Stevens]

Facts are facts.......some people just don't get it.

[Graeme Nattress]

Michael, I'll take you through your section a bit at a time:

"Also when you use a Bayer pattern and you de-bayer in the end up loosing a portion of the true resolution you started with."

Yes, but not as much as you'd think, and other sensor topologies are not immune to this, especially if you value good low aliasing performance. This issue is simply that in a sampled system, which any sensor is, if you try to get it to sample a higher resolution than you have samples for, you will run into aliasing.

" Maybe its Red’s goal to double up to 8K so in the end you have a true 6K output to work with?"

Say we had an 8k M-X sensor, if the lens could keep up with it, I'd expect a measured resolution of around 6.4k. But say I had a monochrome sensor (no colour filter array) you may think I'd get 8k resolution from it. I could, but only at the expense of aliasing. If I filter to avoid such nasties, which are generally much worse in motion than in stills (because the aliasing moves in the opposite direction to the motion in the scene, causing visual more distraction, and you can sometimes hand remove aliasing on a still, but that's hardly practical in motion) and I might get ~7k measured resolution.

"The only problem with that is that if you increase resolution you also potentially increase moire opportunity depending on the methods used." - actually no. The higher the resolution of the sensor, the higher the resolution of the input that is necessary to provoke aliasing. As the world and the lens are staying the same (inputting the same detail into the camera system), but the resolution of the sensor increases, this means that propensity to alias is lower, not higher. By increasing the sensor resolution you're increasing the level of detail the system can cope with without aliasing. This is just basic sampling theory. Now, of course, you may have seen cameras with very high resolution sensors (like the 5D2) that alias very badly, but that's because the full sensor is not read, leaving gaps in the data (line skipping). This has two effects - the OLPF that was set for the pitch of the pixels for sampling all of them for a photograph is now no longer appropriate for the reduced pixel count, and the gaps between the pixels make for the remaining pixels to act more like point samples, leading to greater sensor MTF which means greater propensity to alias. So more resolution on the sensor can lead to aliasing issues, but only if you don't engineer the system properly and thumb your nose at sampling theory.

" Most all high res cameras have a OPL filter (optical low pass filter) on the sensor that basically softens the image."

All competently engineered cameras have an OLPF, yes. The purpose of the filter is not to soften the image, but to remove (filter) detail which the sensor should never see. Such filters are made from bifringent crystal and are necessarily (unlike electronic filters) slow in that to achieve decent attenuation at cutoff, there must be some loss of MTF at the lower frequencies we'd want retained. Increasing the resolution of the sensor will increase MTF at these lower frequencies (because the pixels get smaller) and if all is engineered well you'll get very high MTF at lower spatial frequencies and minimal aliasing.


"Graeme at Red claims this is not the issue with Red and they do not need to add a stronger OLP filter as the pixel size remains the same."

This is basic sampling theory stuff again. The resolution of the system is governed by pixel size (and fill factor) and if they remain constant, the OLPF thickness for the layers of lithium niobate remain constant. If you increase the resolution, you can decrease the thickness of the crystal layers on the filter and hence it gets weaker. As long as the pixel size and crystal thickness are in proportion, all works. Now..... If you go for a really big sensor you reduce the need for the OLPF as you're going to be shooting with a narrower aperture to achieve useful DOF, and may get into diffraction limiting. As you increase the sensor resolution (by decreasing pixel size) you may reduce the need for the OLPF as the lens may limit the incoming MTF sufficiently (especially when combined with the diffraction limiting at typical shooting apertures). Generally though, for an S35 or 35 sized sensor, we're not into diffraction limiting and the lenses are still really good.

"Fair enough, but not all companies follow the same practices and I will have to take his word for it."
In the stills world, generally OLPFs are weaker than we'd allow for with motion, because aliasing is easier to remove (it's just a still), and it's less objectionable (it's not moving contrary to motion). Sigma for instance avoid an OLPF entirely, but images often suffer from aliasing or stair-steps on hard edges. Some people like this.....

" Regardless of this, resolution does have a cap with the human eye and the brains ability to process that image. So while the argument for resolution not adding moire may be true, it still doesn’t change the fact of cognitive resolution by humans which means no benefit to the user in the end."

As noted above, increasing resolution beyond what can be seen leads to higher MTF at the lower frequencies that can be seen, lowers the potential for aliasing and relaxes the constraints on the OLPF - all of which are good things for the image.

"Its best to just avoid Bayer all together or find better ways to arrange patterns to incorporate more green which I think is what most companies are focusing on." - this is a non-sequitur in your argument as it doesn't follow on from the premises above.

"Canon C500, Sony F65 both do not use a standard Bayer pattern" - C500 is utterly standard. F65 rotates the pixels through 45degrees, keeping the exact same ratio of red:blue:green as per standard Bayer. What the F65 does different is it uses 5.6k worth of pixels to produce a 4k image that is claimed inaccurately (unless you're into stair step counting) to come from a 8k sensor. So for 4k they use 17.7mp which is more than the ~9mp a R1 would use to create 4k. This is a good thing (assuming their optical filtering is set optimally to avoid aliasing). That is why Epic uses 5k to produce 4k (or 5k if you want....) from ~14mp. However F65 claims to be 8k which would mean it would have to have ~36mp to do 8k like the R1 does 4k, or 50mp to do 8k the way it does 4k, neither of which it achieves.... Talking just about 4k from the F65 for a moment, what it means by the rotation is that they don't have to demosaic the green, but that means they're extracting 4k of samples as-is (but with lower equivalent fill factor than if they'd had a 4k monochrome sensor like my example above) - which means that the resolution from that 4k sensor is limited to 4k if you accept for aliasing and ~3.5k if decent optical filtering is used.

"and therefore can retain more of their native resolution" - as pointed out at the start, you can do this on any camera if you relax low aliasing requirements. Statements about resolution that don't include aliasing performance are just telling you one side of a (I hope you now see) complex set of performance characteristics. In any case with the F65, you'd have to compare it's measured resolution to either the 5.6k worth of pixels or Sony's 8k sensor claim as "native".

" (not all of it as they are still Bayer pattern but have more green values) . "
Neither have more greens than a standard bayer pattern as the ratio is identical.

"The Arri Alexa starts with a sensor which is much larger than 1920×1080 knowing that the De-Bayer will yield a nice 2K/1080P output." indeed it does, but struggles to achieve an alias-free HD image and struggles to measure HD resolution too (there's detail out at 2k, but it's quite aliased so I'd not really count it as true resolution). Certainly a RED can provide a vastly cleaner (in terms of low aliasing / moire) and higher MTF at 2k.


" Ideally in the future 4K versions of cameras coming out should have a goal of Bayer free 4K output with a 1080p output option." - I don't know why you say "Bayer-free", but 4K measured resolution with low aliasing is certainly a worthy goal. 5k Epic will achieve that, as probably will the 5.6k F65 (not quite sure on aliasing performance there yet).

[Mark Phelan]

Graeme, you have the patience of Job.

[Mike P.]

Graeme, you have the patience of Job.

It's spelled "Gob", Mark. And yes, he does.

[Philipe Ratton]

Graeme, you have the patience of Job.

Amen to that.