"Actual" resolution discussion

[Axel Mertes]

[Mod note: OT posts moved from Jim's RR thread in Recon]

Hi Guys,

we just finished on grading our 10 RED camera 3 hours long concert shoot for german "ASP" band using SpeedGradeDI.

We worked for a 1080p release output (BluRay) and often zoomed in up to 200% (ie. a quarter RED camera image becomes full screen). In that case you essentially NEED to go the 4K route just to resolve acceptable 2K results.

We transcoded R3Ds (frame to frame selected takes) to CineForm RAW at full 4K and graded with that footage, fully realtime, on SpeedGradeDI, with crop & pan & scan to full HD. It was fluent and an amazing workflow experience. Btw, actually impossible to do on Scratch.

We also played our R3Ds from that concert footage on the new Clipster 4K with the RED option installed onto a QuadHD display - NICE. If the RED Rocket delivers the same experience, then NICE. But as of now, no one needs a 4K display for grading - they are simply not good enough for this yet. It makes no sense.

But technically the RED Rocket would not advance our workflow here other than improving transcoding speed to CineForm RAW or Adobe DNG to keep that 4K realtime grading workflow. I don't want to go a step back.

If I really need to go 4K RED with full RED debayered RGB in realtime, I'd choose a Clipster 4K as base platform. This has the power to do it. We've seen it, we've tested it, with our own footage.

Cheers,
Axel

 

[David_Cox]

This is not true. We've tested that the RED format clearly resolves to 3.3K no problem and rendering out to 4k is a good idea because one gets nice transitions rather than harsh pixel for pixel detail.

It depends what colours are in your scene. A bayered 4K (2:1 aspect ratio) image is effectively an overlay of a 2k x 2k green image, a 2k x 1k blue image and a 2k x 1k red image. If you were to point the camera at a grid that had the perfect chequerboard of green, red and blue dots, you could theoretically get a full 4K image. Likewise if you pointed the camera at a pure blue screen, you would probably only get a 2k x 1K image. Most real world scenes would be limited by the 2K green image, although they could resolve to a little higher than this, perhaps 2.3K, because of the greater bias of green in the luminence value of the scene.

If you have seen a more likable image rendered up to 4K as opposed to 2K, this will probably be due to the softness that will be inherent in the up-scaling from bayered 4K to de-bayered 4K. As such this will vary from software to software, as different vendors use different anti-aliasing techniques. I agree that a razor sharp RED image does need a little help though!

My point about recommending 2K over 4K as a post production file format is that there is very, very little visual difference (if any at all) in a properly debayered and rescaled 4K RED image to 2K post work flow, than a 4K post work flow, but the difference in time taken and so the cost to the end client is quite dramatic. I think it is more important to spend the clients money on what makes the end film significantly better rather than what might be mathematically better, but very hard to see (especially when you add in how and where the final film is viewed)

dc

 

[Jeff Kilgroe]

It depends what colours are in your scene. A bayered 4K (2:1 aspect ratio) image is effectively an overlay of a 2k x 2k green image, a 1k x 1k blue image and a 1k x 1k red image. If you were to point the camera at a grid that had the perfect chequerboard of green, red and blue dots, you could theoretically get a full 4K image. Likewise if you pointed the camera at a pure blue screen, you would probably only get a 1K image. Most real world scenes would be limited by the 2K green image, although they could resolve to a little higher than this, perhaps 2.3K, because of the greater bias of green in the luminence value of the scene.

That's a load of bunk. What you're saying only somewhat applies to chroma resolution, but completely discounts luma resolution. A proper debayer process also takes a lot more into account than simply registering a 1k x 1k raster for red and blue and 2k x 2k for green. And actually your numbers are off based on the 2:1 aspect you're using. 4K 2:1 would actually have the equivalent of 2K x 1K for both red and blue. With the Mysterium and most other CMOS sensors, all photosites record luma information, even if no light within their affinity range is registered. Yes, it's true that you will see a lower apparent resolution if you shoot a scene that is 100% blue, as an example, but it's not going to completely discount 75% of the sensor resolution and luma res is still going to register around the 3K mark.

My point about recommending 2K over 4K as a post production file format is that there is very, very little visual difference (if any at all) in a properly debayered and rescaled 4K RED image to 2K post work flow, than a 4K post work flow,

That's entirely subjective. But I'm seeing very visibly noticeable and mathematically measurable advantages to a full 4K debayer vs. a 2K debayer.

but the difference in time taken and so the cost to the end client is quite dramatic. I think it is more important to spend the clients money on what makes the end film significantly better rather than what might be mathematically better, but very hard to see (especially when you add in how and where the final film is viewed)

Cost is an issue, but I think 4K post is about to go mainstream. With the ease of working with 2K these days, and even 4K, clients are going to start to demand a full 4K post process and prices are dropping fast. Economy solutions like Adobe CS4 can push 4K source material through the entire pipeline and already provide more complete and capable 4K solutions than what many of us used for 2K/1080p a few years back. Any post house that is trying to be a true industry player is already missing the boat if they're not providing 4K post to RED clients.

And to say that 4K post offers no visible difference over 2K post is the same thing as saying we shouldn't even bother with more than 1366x768 resolution for most post workflows because 35mm film distribution prints rarely exhibit more resolution than that and that's the native resolution of approximately 85% of all HDTV displays sold over the past 5 years.

Now, I would agree that if we were to post at 3K, there would be really no visible advantage to moving to 4K for RED originated footage. However, no one works in 3K because they just don't.

And with 5K Mysterium-X cameras releasing by the end of this year (or at least most indicators point to that), anyone not providing a full 4K post pipeline is going to be overlooked by those buying these new cameras. I can handle workable 4K post solutions in my home office, hardware like the Red Rocket will only improve this situation and the speed at which I can work. The only piece of the puzzle that's truly missing is a proper and affordable monitoring solution and that's just around the corner - I expect to see some truly reasonable solutions early next year at CES and NAB.

 

[Steven Caesare]

It depends what colours are in your scene. A bayered 4K (2:1 aspect ratio) image is effectively an overlay of a 2k x 2k green image, a 1k x 1k blue image and a 1k x 1k red image. If you were to point the camera at a grid that had the perfect chequerboard of green, red and blue dots, you could theoretically get a full 4K image. Likewise if you pointed the camera at a pure blue screen, you would probably only get a 1K image. Most real world scenes would be limited by the 2K green image, although they could resolve to a little higher than this, perhaps 2.3K, because of the greater bias of green in the luminence value of the scene.

If you have seen a more likable image rendered up to 4K as opposed to 2K, this will probably be due to the softness that will be inherent in the up-scaling from bayered 4K to de-bayered 4K. As such this will vary from software to software, as different vendors use different anti-aliasing techniques. I agree that a razor sharp RED image does need a little help though!

My point about recommending 2K over 4K as a post production file format is that there is very, very little visual difference (if any at all) in a properly debayered and rescaled 4K RED image to 2K post work flow, than a 4K post work flow, but the difference in time taken and so the cost to the end client is quite dramatic. I think it is more important to spend the clients money on what makes the end film significantly better rather than what might be mathematically better, but very hard to see (especially when you add in how and where the final film is viewed)

dc

I don't believe it's quite that simplistic. I believe that the algorithm is capable of taking account of the spectral energy of light hitting adjacent pixels in scaling the individual color channels to the full size.

Graeme has stated something on the order of 78% of the full pixel resolution can be demonstrated without significant artifacting.... and that's after the OLPF.

-Steve

 

[David_Cox]

I think I'm being mis-interpreted. What I am talking about is value to the end client, not satisfying the engineers. I did have typo's on my second response though that I have amended regarding 1k.

What RED has achieved is to be able to provide high quality photography at a much reduced price. So there is an inherent demand for value by anyone who uses it. That value is a balance between image quality and cost.

Our workflow is to debayer to 4K and decide at that point what resolution, in an uncompressed, non bayered world, to proceed in. We use the highest quality rescalers that make sure we don't encounter aliasing due to resolution changes. I have yet to see a project shot on RED ONE where the benefits of proceeding at 4K outweigh the additonal cost in final quality, caused by having to finish the creative process sooner in order to allow more time for 4K deliverables over 2K, HD or whatever.

We could post produce at 4K, 8K, whatever K. If a client demands a resolution, we'll do it. But most demand "production value" and that means making the most of the budget and deadline restrictions and not concentrating on something that most people will never see. This way, we deliver to the end client the value that the RED ONE camera uniquely provides.

DC

 

[Steven Caesare]

Despite your corrections, you still say:

"Most real world scenes would be limited by the 2K green image, although they could resolve to a little higher than this, perhaps 2.3K, because of the greater bias of green in the luminence value of the scene."

That has been demonstrated to be untrue.

You, or your client, may not feel you need it for your final delivery, but there is data there in excess of what you quote.

 

[Graeme Nattress]

http://www.cinematography.net/natress-red-res.html

The MTF plots show detail all the way out to 4K, and significant detail beyond 3k. That is because all photosites are used to help compute luma, not just the green ones.

I'd be very keen to see a scale of the 4k zone plate to 2k that does not induce any extra aliasing...

Graeme