Scarlet Dragon Resolution vs. compression decisions

[Adam LaBrie]

Hey guys, Ive been super happy shooting with my scarlet dragon these past few months.

I'm gearing up for a boxing short film (3 night shoots) shooting on super speeds.

I want to shoot in 5k to get truer focal length and more pixels is always better, but the lowest compression scarlet dragon allows is 8:1 at 5k.

I plan on rating at 800 and part of me wants to go with 4k just to be able to shoot at 5:1, which I know will help with noise and artifacting at higher iso. Any thoughts are appreciated, thanks guys (and gals)!

 

[sam karr]

I was checking this thread: http://www.reduser.net/forum/showthread.php?117974-Any-Scarlet-Dragons-out-there/page11. There is some R3D comparison files, MX vs dragon at 8:1. The compression looks better on the dragon, as some people already reported. MX noise is really more messy from what i see. I can't quantify the difference, and hope someone will do, but i notice a difference in favor of the dragon. So 8:1 on dragon is not a limitation. It's probably true that it is the equivalent of 5:1 on epic, as I read somewhere else.

 

[Tommaso Alvisi]

Adam, I'd say go for 5K @ 8:1 absolutely, you will downsample that to 4K or 1080P so...and 8:1 is not bad on dragon...

 

[Akin A]

I'm also interested in knowing more about what has a bigger effect image quality and where your threshold for acceptability ends on Scarlet Dragon: high resolution/high compression vs lower resolution/low compression.

From these examples, I was surprised to see how unimpressive 2K WS 7:1 120fps and 3K WS 10:1 100fps appear, compared to 5K WS 16:1 60fps and 4K WS 13:1 75fps.

Is the reason for the quality jump soley the effect of oversampling? Is oversampling that much more powerful than low compression?

Or is part of the reason for 2K/3K HFR being so much worse because faster shutter speeds were required, therefore a larger aperture, which caused softness at the lens level? Or is it because the photosites are so tiny in this 6K sensor, causing the quality of light captured to hindered?
Or all of the above?

 

[Anton Shavlik]

I have been able to get very nice 2k images out of my scarlet dragon. I think you saw my video in the other thread, but I'll post it again here.

2k HD 8:1 compression, 180 degree shutter @ 120fps. Zeiss ZE 50mm 1.4 @F4 (IIRC). Developed at 100iso in Resolve. I blackshaded just before shooting.

I feel very strongly that noise / softness at lower resolutions is a redcode issue and the only solution is to get an immaculate exposure, I used these same lenses on my blackmagic 2.5k and they were ultra sharp. I shoot the same settings but at 5k and the noise/softness is the same if you zoom into a 2k extraction.

 

[Akin A]

I have been able to get very nice 2k images out of my scarlet dragon. I think you saw my video in the other thread, but I'll post it again here.

2k HD 8:1 compression, 180 degree shutter @ 120fps. Zeiss ZE 50mm 1.4 @F4 (IIRC). Developed at 100iso in Resolve. I blackshaded just before shooting.

I feel very strongly that noise / softness at lower resolutions is a redcode issue and the only solution is to get an immaculate exposure, I used these same lenses on my blackmagic 2.5k and they were ultra sharp. I shoot the same settings but at 5k and the noise/softness is the same if you zoom into a 2k extraction.

Thanks Anton. Yeah, I remember seeing this.
I see the small embedded video, but I can't seem to watch on vimeo or make it full screen to really judge the quality. From the small embed, it looks good though. Was the original image so noisy that you had to set ISO to 100 to clean it or was that just your preference?

I'm also coming from a BMCC 2.5K and I was thinking that a similar crop of 2.5K or 2K (compared to Pocket cam) on Scarlet Dragon would have at least as much detail. I know the codecs/compression/bit-depth are different, so it's not apples to apples.

Any thoughts on the technical reason for the quality drop in the other videos I found (besides the lens chromatic aberration)? What do you feel about the relationship between resolution oversampling w/ higher resolution at higher compression vs lower resolution w/ low compression in regards to image quality? I would think that at some threshold, the image quality would be the same. It would be nice to identify that point.

Thanks for pointing out that shooting 5K and cropping 2K in post looks the same as shooting cropped 2K. That was going to be my next question.

It would be great if Scarlet Dragon could at some point take advantage of Mini-Mags to allow lower compression.

 

[Björn Benckert]

According to my findings resolution is to be preferred over lower compression/lower resolution. So I would say 5k at highest possible bit rate will look better than 4k at highest possible bit rate after downsampling. It's not only the downsampling that helps but you also use more resolution / larger image circle of the lens. So I always recommend to use as much of the sensor as possible. If mastering in wide screen it's off course better if possible to shoot widescreen and lower compression than full frame with higher compression... even though some would argue that they would want the FF for reframing and that they find the difference in compression to not be an issue.

 

[Michael Tiemann]

Thanks Anton. Yeah, I remember seeing this.
I see the small embedded video, but I can't seem to watch on vimeo or make it full screen to really judge the quality. From the small embed, it looks good though. Was the original image so noisy that you had to set ISO to 100 to clean it or was that just your preference?

Clue #1: know your finish. If you know you are going to finish at 1920x1080, then to first approximation, debayering anything more than 2880x1620 should be overkill (see OLPF comments below). And anything less is going to start showing weakness.

I'm also coming from a BMCC 2.5K and I was thinking that a similar crop of 2.5K or 2K (compared to Pocket cam) on Scarlet Dragon would have at least as much detail.

The BMCC famously has no OLPF, which helps it make images that can edges very sharp. The downside of omitting the OLPF is that is makes the sensor much, much more sensitive to moire. For me, I get moire patterns so frequently and consistently that I'm about ready to sell my BMCC. Resolve can sometimes smooth out these patterns, but I've found enough examples where they cannot be fixed that I've just given up. I've never seen moire with my DRAGONs, but it comes at a cost--edges are nothing like the BMCC, though they can be made so by sharpening and/or scaling down. To get really sharp edges, one might use another factor of 1.5x, so now we're in 4K capture territory for a clean, sharp 2K finish.

I know the codecs/compression/bit-depth are different, so it's not apples to apples.

BMCC is uncompressed RAW, while RED is wavelet-compressed RAW, so there's definitely a cost there, too. The extra smoothing that the OLPF does definitely give the wavelets license to lock in that smoothness. When using low compression rations, I believe that the wavelet smoothing and the OLPF smoothing don't really accumulate too much, so the extra resolution used for sharpening the OLPF goes a long way toward properly sharpening what the wavelets might also meld together. But at higher compression ratios we get solidly into information theory and perception theory. And only experience teaches us which scenes will tolerate what levels of compress for a given finish.

Any thoughts on the technical reason for the quality drop in the other videos I found (besides the lens chromatic aberration)? What do you feel about the relationship between resolution oversampling w/ higher resolution at higher compression vs lower resolution w/ low compression in regards to image quality? I would think that at some threshold, the image quality would be the same. It would be nice to identify that point.

I think it just comes down to math. 5K @ 16:1 is the same bitrate as 2.5K @ 4:1. Oh...wait: the test you showed kept upping the FPS, so their 2.5K was shot at 7:1. That's like 5K@28:1, which is literally off the scale of what DRAGON supports. Thus, it is no wonder that as the test progresses, things look progressively worse. First, because the actual bitstream is being starved as the framerate speeds up, and second because 2.5K is not a happy place for a 2K finish.

Thanks for pointing out that shooting 5K and cropping 2K in post looks the same as shooting cropped 2K. That was going to be my next question.

It would be great if Scarlet Dragon could at some point take advantage of Mini-Mags to allow lower compression.

Scarlet cannot take advantage of the mini-mags because it doesn't have the CPU power necessary to crunch the bits fast enough to reach those lower compression ratios. That's another thing EPIC DRAGON buys you.

 

[Akin A]

Michael, thanks for the explanations! Much clearer now!

 

[Anton Shavlik]

Full debayer 5k on scarlet dragon at the lowest compression available is the best look, without a doubt, from my limited experience with the camera. I shoot 5:1 @4k a lot too, especially when I'm using green screen for a 1080p finish.

I shot that 1080p 120fps clip at 250 iso in camera and used some sharpening / pushed the iso lower because I wanted to see just how good the images under ideal settings can be, basically to decide if I would rather rent a FS700 for projects that called for high frame rate.

 

[Lee Saxon]

Is the reason for the quality jump soley the effect of oversampling? Is oversampling that much more powerful than low compression?

Or is part of the reason for 2K/3K HFR being so much worse because faster shutter speeds were required, therefore a larger aperture, which caused softness at the lens level? Or is it because the photosites are so tiny in this 6K sensor, causing the quality of light captured to hindered?
Or all of the above?

The larger aperture could make a difference, though anyone who can afford these cameras shouldn't be futzing around with any lens poor enough to make that difference a problematic one. It's not the photosites; Sony's 36mp FF35 sensor has smaller ones and that sensor's output is considered amongst the best currently available (also, lower resolutions on Dragon are just crops of the sensor, there's none of the pixel binning or line skipping or any of the garbage VDSLRs do, so any image quality issues caused by sensor properties would appear at all resolutions, not just 2K/3K).

It's 99.99999% the downsampling. You're working with massively more information, so you're definitely going to end up with a much more detailed image.

What I find strange is people using these tests (which I call "downsampling tests") - 5K footage downsampled to 2K and played beside 2K footage, for instance - as evidence that there is something wrong with Dragon or Redcode at low resolutions. It's not. In fact, 5K has such a significant natural advantage in that situation that for 2K to compare well would mean that there was something wrong with Dragon or Redcode at high resolutions (or that they'd excessive compression on the 5K footage)!! Additionally, the quality of your NLE's resampling algorithms becomes a factor in such tests.

To assess whether there was a problem at low resolutions, or whether it's a good idea to lower resolution so you can use a better compression level, you'd need to do a "cropping test" - 5K footage cropped to 2K and played beside 2K footage (ie, view both at 100%, unresampled).

It's a matter of perspective. Most of the time you shoot higher resolutions to downsample, not crop, so the "downsampling test" can be seen as a more useful indication of what to expect in many "real world" situations. But personally, I'm more interested in the "cropping test" and wish more people - especially the ones trying to determine if there's some Redcode issue with lower resolutions - would do it.

 

[Akin A]

Thanks for the input, Lee.

I was considering shooting 5K to crop in post to 4K and reframe and simulate camera movement. I did that all the time w/ BMCC, shooting 2.5K and cropping 2K to reframe w/ movement. It might sound pointless, but it's good for when I'm shooting solo in my studio on a tripod w/ no camera operator.

Sounds like that probably wouldn't look too great at 5K cropped to 4K on Scarlet if I can't oversample. Anybody do this?

 

[Michael Tiemann]

Thanks for the input, Lee.

I was considering shooting 5K to crop in post to 4K and reframe and simulate camera movement. I did that all the time w/ BMCC, shooting 2.5K and cropping 2K to reframe w/ movement. It might sound pointless, but it's good for when I'm shooting solo in my studio on a tripod w/ no camera operator.

Sounds like that probably wouldn't look too great at 5K cropped to 4K on Scarlet if I can't oversample. Anybody do this?

What's your delivery medium? If you are aiming for a 70' wide projection screen, you might get caught out: http://www.red.com/learn/red-101/eyesight-4k-resolution-viewing

If you are aiming for a 85" UHD-TV, then anybody sitting 8.5' away from the screen won't be able to resolve more than 3K pixels anyway [ibid]. (If you are shooting for 65" UHD-TV, then that shrinks to 6.5', obviously.)

If you've been happy with a 2K crop from the BMD for HD, then you should be happy with a 4K crop from Scarlet for UHD, if you sharpen a bit (to take care of the strong OLPF).

 

[Akin A]

Which leads to my next question about shooting 6K stills.
For those of you who shoot 6K stills on Scarlet Dragon (I plan to), do you find yourself needing to add extra sharpness in post?

I guess you can get away w/ that more on stills than you would with motion.

 

[Akin A]

What's your delivery medium? If you are aiming for a 70' wide projection screen, you might get caught out: http://www.red.com/learn/red-101/eyesight-4k-resolution-viewing

If you are aiming for a 85" UHD-TV, then anybody sitting 8.5' away from the screen won't be able to resolve more than 3K pixels anyway [ibid]. (If you are shooting for 65" UHD-TV, then that shrinks to 6.5', obviously.)

If you've been happy with a 2K crop from the BMD for HD, then you should be happy with a 4K crop from Scarlet for UHD, if you sharpen a bit (to take care of the strong OLPF).

Unknown delivery medium at the moment, just thinking about the future.
I've always hated the idea of sharpening video in post. I'll just have to get used to it if it's not that bad in reality.
Thanks for the link.

 

[Michael Tiemann]

Which leads to my next question about shooting 6K stills.
For those of you who shoot 6K stills on Scarlet Dragon (I plan to), do you find yourself needing to add extra sharpness in post?

I guess you can get away w/ that more on stills than you would with motion.

SCARLET DRAGON is the same as EPIC DRAGON when it comes to stills. EPIC DRAGON is the best sensor ever measured by DxO: http://www.dxomark.com/Reviews/RED-...core-Epic-Dragon-takes-first-place-101-points

Secondly, the DRAGON's sensor is 30.7mm wide and 15.8mm high. If you are shooting Leica Summilux primes that do resolve 200 lp/mm, the sensor is right about at the limit of what the lens can deliver. And not all such lenses deliver 200 lp/mm across the whole image frame. But the vast majority of lenses peak at 140 lp/mm or less in the center (stopped down) and more typically deliver 100 lp/mm or less when wide open, toward the edges, or both. Which means that the DRAGON's 30.7mm sensor will collect 3.5K of resolution in the best of circumstances and probably less than 3K of resolution more typically. So the sensor is not going to be your problem when you shoot 6K stills.

 

[francisco santos]

Unknown delivery medium at the moment, just thinking about the future.
I've always hated the idea of sharpening video in post. I'll just have to get used to it if it's not that bad in reality.
Thanks for the link.

learning proper denoising and sharpening too my red footage to the next level.

 

[Robert Ruffo New]

It's mostly lens field of view. You are zoomed out of any imperfections your lens and even your filters might have. You are also zoomed out of compression issues, noise and so on.

 

[Gavin Greenwalt]

To answer the simple question of "Resolution vs Compression".

If I had the choice between: A) 100MB/s 6k and B) 100MB/s 4k I would pick 100MB/s 6k all else being equal. The reason for that is that compression algorithms by their very nature are "Smart" they look at high detail regions and throw mb/s at them. They look at low detail regions and save the mb/s for the higher detail regions.

A 4k image will in any given region be at most 4k. A 6k image in any given region will be at most 6k. High compression 6k will most likely be on average less than 6k measurable resolution but if the image has enough low detail regions the 6k regions will probably approach their hypothetical maximum with minimal compression which will be at least higher than 4k. To take an extreme example an uncompressed 720p definition RAW video is 21MB/s 12:1 4k is also 21MB/s and you would be hard pressed to find 4k video that looks softer than 720p in most situations (short of a grassy field).

If we were just filming pure noise (or a field of grass) then I would say it's probably a wash, 6 one half a dozen another if you're compressing the image neither the 6k nor the 4k will meet their full potential and you'll probably end up with a 3k image anyway just because compression isn't magic. You can get about 2:1 compression on random noise but if you have pure noise and your compression is at 6:1 you're just throwing away 60% of the actual image detail.

The other reason I would favor 6k over 4k is due to lenses. A lens that gives you 80 lp/mm will give you a maximum resolution on a 20mm wide (4k) portion of the senor of 20mm*80lp/mm = 1600 *2 = 3.2k. The same lens on the 6k portion of the sensor would deliver 30mm * 80lp/mm = 2400 lp * 2 = 4.8k resolution. So even if your compression was perfect on the 4k in that example your lens on the smaller sensor crop wouldn't.

 

[Charles Bergquist]

Are the Scarlet Dragon compression breakdowns somewhere?

I can't seem to find them with Mini-Mag 512.