Raven Redcode Ratios

[Nathan DuMoulin]

I'm new to the Red ecosystem, and have pre-ordered a Raven. I've downloaded some clips from other forum members (thanks!), most of it being 4K sensor crops of the larger Dragon, at medium compression (5:1, 8:1 sort of thing).

My curiosity falls into two categories:

1) 4K is at 120fps, which is a big deal for my work (sports). It is however, at 13:1. I've read that anything above 10:1 starts to fall apart, while others have said that they can't see any problems. Can anyone offer any samples or give their personal experience regarding shooting slow motion at high levels of compression? One forum member was kind enough to post a skier in the snow at 120fps, but it was overexposed, contained less than 1% shadows, and had light snow falling that was obscuring shadow detail. I appreciated the user for sharing it, but unfortunately it didn't really help solve my curiosity.

2) Still speaking in terms of 120fps, I'm assuming that if you can do 4K at 13:1, it's likely also possible to do 3K at 10:1 (estimated) and 2K at 7:1 (also a guess). I'd imagine these all max out the datarate. In terms of acquiring optimal quality, which would yield a cleaner picture, 4K at 13:1 down-sampled to 3K, or shooting natively at 3K at 10:1? Does resolution trump, or does the cleaner compression?

Thanks!

 

[Michael Tiemann]

I find 4k @13:1 to be really noisy in the shadows. And the noise is not pretty. Here is a random R3D with some shadows so you can see how the noise looks like. It's 4k @13:1 60fps on scarlet dragon.
https://www.dropbox.com/s/8g57xo6ic2074hx/A001_C018_0318RQ_001.R3D?dl=0

Of course in a less extreme shot, with good lighting, you can make 13:1 usable.

I'll lay down a dollar that that shot would look just as terrible if shot at 8:1 or even 4:1. It's missing the 10 stops between "too bright" and "too dark" and there's nowhere for REDCODE to actually encode anything interesting.

 

[sam karr]

I'll lay down a dollar that that shot would look just as terrible if shot at 8:1 or even 4:1. It's missing the 10 stops between "too bright" and "too dark" and there's nowhere for REDCODE to actually encode anything interesting.

Well the purpose is to show the grain structure @13:1. I said before that with good lighting you can have some nice images. But it's also interesting to see what kind of noise structure you can expect in some areas @13:1.
If shot @8:1 or 4:1 you would have some less blotchy noise, and even super nice grain @4:1.

 

[Robino_J]

The only time 13:1 was unacceptable for me was in a forrest. The detail was just too much for the codec. Generally I have not seen major issues.

 

[Michael Tiemann]

Well the purpose is to show the grain structure @13:1. I said before that with good lighting you can have some nice images. But it's also interesting to see what kind of noise structure you can expect in some areas.

I disagree. If you shoot a lens cap, you get 100% noise. That's not interesting. What is interesting is when noise becomes the reason that an otherwise well-executed shot is no longer good enough. In this particular case, once you grade out all the noise, there's nothing left, because there was really nothing to begin with.

I went back through my various HFR projects, and though I thought I'd shot some above 10:1, the highest I could find is 9:1, and most are 8:1. That's because I mostly use my EPIC DRAGON, not my SCARLET DRAGON.

I think that Akin A (http://www.reduser.net/forum/member.php?176005-Akin-A) did some SCARLET DRAGON HFR tests with commentary about higher compression ratios.

 

[sam karr]

I disagree. If you shoot a lens cap, you get 100% noise. That's not interesting. What is interesting is when noise becomes the reason that an otherwise well-executed shot is no longer good enough. In this particular case, once you grade out all the noise, there's nothing left, because there was really nothing to begin with.

I went back through my various HFR projects, and though I thought I'd shot some above 10:1, the highest I could find is 9:1, and most are 8:1. That's because I mostly use my EPIC DRAGON, not my SCARLET DRAGON.

I think that Akin A (http://www.reduser.net/forum/member.php?176005-Akin-A) did some SCARLET DRAGON HFR tests with commentary about higher compression ratios.

Ok I deleted my file as you're not happy with it.
So you tell me you never shot above 10:1. I guess of course you have less trouble with compression.
I thought it would be interesting to see how the grain looks. In each of my 13:1 shots, even when well lit, any shadow part shows this kind of blotchy noise.
So yeah let's post some overexposed footages so every future raven owner will be happy not seeing any noise and think they will shoot hyper clean 120fps....

 

[Michael Tiemann]

[...]In each of my 13:1 shots, even when well lit, any shadow part shows this kind of blotchy noise.

One of those shots would be interesting, especially if it shows where 13:1 holds well, where it falls apart, and whether the compromise seems workable. Everything (mostly) is a compromise, after all. It's just that some compromises are workable, others less so.

So yeah let's post some overexposed footages so every future raven owner will be happy not seeing any noise and think they will shoot hyper clean 120fps....

^ Not what I said.

 

[Marcus Friedlander]

Link to Akin's test

http://www.reduser.net/forum/showthread.php?125466-Scarlet-Dragon-Slow-Motion-Test-Knife

 

[Christopher-lee dos Santo]

Take a look at two little slow motion test videos done with the Scarlet Dragon. The first two shots of the BMW video are 2kWS upscaled to a 4k timeline at 150fps. Really trying to see how usable the 2k footage was. The rest of the slow motion are 5kws at 60fps at 16:1.

The other video has titles which show what lens, resolution and redcode was used.

Bare in mind this is the bigger Dragon Sensor with 19megapixels, so not directly comparable to the 8 megapixel sensor in Raven, but should give you some idea.

1st video - https://vimeo.com/115074032

2nd video - https://vimeo.com/111890797

 

[Nathan DuMoulin]

Link to Akin's test

Take a look at two little slow motion test videos done with the Scarlet Dragon.

Although it's obviously subjective, 2K is very debatable, and seems to be passable in limited situations, if at all (I think the BMW clip may be a pleasant exception). 3K seems improved, but still likely limited to 1080p delivery. The 4K at 76fps 13:1 actually looked far better than expected, even with shitty youtube compression. Nice to also see an example with controlled studio lighting.

These were both hugely helpful. Thank you.

Bare in mind this is the bigger Dragon Sensor with 19megapixels, so not directly comparable to the 8 megapixel sensor in Raven, but should give you some idea.

I've heard a few people make a similar statement, and I must admit that I'm slightly confused. Isn't the full size of the sensor relative to the area being cropped irrelevant? Meaning, isn't a 3K crop on a 4K Dragon sensor going to yield the same result as a 3K crop on an 8K Dragon sensor? It's processing equal amounts of a Dragon sensor, so why would the result be any different?

Thanks again everyone!

 

[sam karr]

I've heard a few people make a similar statement, and I must admit that I'm slightly confused. Isn't the full size of the sensor relative to the area being cropped irrelevant? Meaning, isn't a 3K crop on a 4K Dragon sensor going to yield the same result as a 3K crop on an 8K Dragon sensor? It's processing equal amounts of a Dragon sensor, so why would the result be any different?

Thanks again everyone!

I think he was referring to the fact that if you shoot @5k and output/downscale @2k for example, as there is 50-60% more pixels @5k than 4k, the grain will be smaller at the end, and the noise less distracting than if you shot @4k and output/downscale @2k. But shooting 4k on raven and 4k on weapon will give the same result for the same compression rate.

 

[Michael Tiemann]

I think he was referring to the fact that if you shoot @5k and output/downscale @2k for example, as there is 50-60% more pixels @5k than 4k, the grain will be smaller at the end, and the noise less distracting than if you shot @4k and output/downscale @2k. But shooting 4k on raven and 4k on weapon will give the same result for the same compression rate.

Yes, and...there's probably also a point where downscaling follows the laws of diminishing returns. From what I've read, good math will largely cure bayer pattern noise when scaling down 66% (such as 6K->4K) and virtually any math will cure the bayer pattern when scaling down 50% (such as 4K -> 2K). Thus, downscaling 5K to 4K vs. straight 4K provides a percentage improvement of MTF across some range. Downscaling 5K->3K is a 60% downscale, comfortably between 50% and 66%. 4K->3K is a 75% downscale, thus just outside the 66% number, but not by much. Both 5K and 4K can be "perfectly" debayered when scaling down to 2K, so the larger sensor doesn't help in that regard. It may help clean up compression noise that's larger than the bayer pattern, but that takes us straight to the question of "is it better to scale then compress, or better to compress then scale". Many have found that the answer is very scene specific.

 

[Darius Lyles]

I haven't shot anything at 13:1 but from the look of these examples if you have enough light then 120fps should be just fine on the Raven. Which means that 100fps and lower should be great.

 

[Nathan DuMoulin]

I have to say that the HFR compressions on Weapon are fantastic and it is the one feature that I am envious of.
We are currently working on a BBC Earth special destined to the 40m x 8m Orbi Screen in Japan http://www.red.com/news/bbc-and-sega-develop-orbi . The maximum recommended compression for this very demanding 6K WS format is 12:1 which seems quite high. I know they have spent a lot of time, effort and money doing their homework for these screens and if that is acceptable for this project I would guess that this compression is acceptable for most other purposes.

Here's a quote from a post made over in the Weapon sub forum.

If 12:1 is passable for BBC Earth, then it's sure as hell passable for my work.

 

[Michael Tiemann]

Here's a quote from a post made over in the Weapon sub forum.

If 12:1 is passable for BBC Earth, then it's sure as hell passable for my work.

Indeed, but...if you want to compare apples and apples, 6K origin vs. 4K origin is 2.25x more pixel information. If compression decimates information content, then 6K reduced by 12:1 compression should be comparable with 4K reduced by 5:1 compression. Obviously some images are more compressible than others, and some lenses resolve 4K over an image circle larger than 20.48mm, so there are reasons why a 6K image compressed at 12:1 will still look better than 4K compressed at 5:1. But the data rates, and thus theoretical information content, are comparable.

 

[Nathan DuMoulin]

Indeed, but...if you want to compare apples and apples, 6K origin vs. 4K origin is 2.25x more pixel information. If compression decimates information content, then 6K reduced by 12:1 compression should be comparable with 4K reduced by 5:1 compression. Obviously some images are more compressible than others, and some lenses resolve 4K over an image circle larger than 20.48mm, so there are reasons why a 6K image compressed at 12:1 will still look better than 4K compressed at 5:1. But the data rates, and thus theoretical information content, are comparable.

Very good point! Thanks for putting that into perspective.

I'm thinking 90-100fps at around 10:1 will be the sweet spot for slow motion for this camera, while 120fps or higher will be reserved for special circumstances.

Which leads me to wonder, will pushing the compression of Raven be more beneficial than say, using Twixtor to interpolate additional frames? My understanding is that Twixtor is not very effective with 60fps or lower, but get's to be decent at 100fps or higher.

Does anyone have any experience with this, or will it be entirely situational?

 

[Fabio Lanzone]

Which leads me to wonder, will pushing the compression of Raven be more beneficial than say, using Twixtor to interpolate additional frames? My understanding is that Twixtor is not very effective with 60fps or lower, but get's to be decent at 100fps or higher.

Does anyone have any experience with this, or will it be entirely situational?

Did a few times with epic 5k at 96fps, doubling it to 192fps. It does a really good job, but you have to keep in mind twixtor rules to avoid artefacts.

 

[Daniel Reed]

The fact that Raven can do 3:1 compression at 4.5K WS/2.4:1 is sure awesome!

in terms of HFR, if the scene has moderate complexity, 60fps at 8:1 may be a good HS target to start at.

with that in mind, Raven is the new Bolex for the modern aspiring filmmaker.
Treat it as such and I suspect most will be happy.

If you're expecting super slow motion, either plan to do low complexity scenes or expect to be disappointed.