Can we discuss REDCODE and 8K?

[Anthony Berenato Jr]

The most challenging thought that looms over my head when considering the jump to Weapon Helium is data management. I love the thought of having an 8K output, alongside all the benefits that come with it... But I'm still wrapping my head around how to manage projects that don't need 8K or even 6K capture. I want to take advantage of the full FOV of my sensor and lens without the increase in HDD space - especially if we're talking solid state.

So here is my thought / question:

Unless RED has some secrets up their sleeve regarding pixel binning and resolution (this would be really exiting if so), I'd love to know if shooting an 8K R3D at a less data-heavy REDCODE rate will net a similar / comparable end-result image in terms of clarity and noise to our current Dragon line at 6K?

Is there a tool or chart that may soon be available to let me compare data rates between the two cams to help find a happy medium?

What do you guys think? I'd love to wait for the new Helium-optimized capture process, but just wondering if anyone has any tests they may have to do a comparison at higher REDCODE rates.

 

[Patrick Tresch]

6k Dragon X 2 = 8k Helium Data rates.

I'm sure there is no R3D coding that can give you 6k 8:1 compression the same result as 8k 16:1 compression.

I bet the 8k S35 downscaled to 4k Prores will be the most used codec from any Epic-W camera.

Pat

 

[Terry VerHaar]

6k Dragon X 2 = 8k Helium Data rates.

I'm sure there is no R3D coding that can give you 6k 8:1 compression the same result as 8k 16:1 compression.

Could you elaborate on this? Not clear on your contention here and where it comes from. Thanks.

 

[Christoffer Glans]

6k Dragon X 2 = 8k Helium Data rates.

I'm sure there is no R3D coding that can give you 6k 8:1 compression the same result as 8k 16:1 compression.

I bet the 8k S35 downscaled to 4k Prores will be the most used codec from any Epic-W camera.

Pat

How did you arrive at 2 x 6K Dragon?

 

[Andy Roberts]

It's just math. 8K is almost twice as many pixels per frame as 6K.

Dragon at 6K FF Resolution = 6144x3160 (19.42 megapixels)

Helium at 8K FF Resolution = 8192x4320 (35.39 megapixels)

Helium at 6K (same exact pixel dimensions) is 1.37x crop relative to Dragon at 6K.

 

[Terry VerHaar]

Side question from a DSMC1 owner who has been ignoring the ProRes recording facts. Is it possible to record ONLY ProRes on a DSMC2 camera; no R3D?

 

[Brendan H. Banks]

Side question from a DSMC1 owner who has been ignoring the ProRes recording facts. Is it possible to record ONLY ProRes on a DSMC2 camera; no R3D?

Yes! All DSMC2 cameras allow this, but each has its own types of recording formats. The EpicW will do 4k ProRes HQ up to 30 (it may be able to do additional things, but this is the only one confirmed). I'll be using the 4k ProRes for a ton of clients who want a more Alexa-like workflow and don't use raw.

 

[James Miller 72]

The thing is as R3D works in wavelet based compression. If you take one of the available Helium 8k R3D's or a native 30mp plus photo and compress both at different JPEG2000 values you can easily reduce the 8k frame to the same apparent resolution and file size in terms of macroblocking and feel. So I can get very close results by increasing compression. So if there is only slight movement then a higher compression can be used but fast movement at low frame rates will need much higher values. Generally speaking I think you could get away with about 1/3 more media but many times come very close as if shooting on the 6k.

 

[Mike Finegan]

Will prores output other than 4K be supported and it is pixel binned or interpolated from full frame 8K capture?
For example 4.5K prores 4:2:2 output from 8K full frame would be very nice format to work with - a little room for framing or motion compensation but dense and sharp (depending on algorithm used). Can anyone confirm prores on Epic W is actually real time interpolation from 8K full frame? Or is it pixel binning?

 

[Andy Roberts]

4K ProRes is the max... at least that is all that is confirmed.

 

[Chad Lancaster]

Why compromise from the beginning? Hard drives are cheap, get some 10TB ones and shoot at minimum compression. You can always lessen the quality of the image in post but you can't get it back if you start at a compromised position.

 

[Blair S. Paulsen]

The thing is as R3D works in wavelet based compression. If you take one of the available Helium 8k R3D's or a native 30mp plus photo and compress both at different JPEG2000 values you can easily reduce the 8k frame to the same apparent resolution and file size in terms of macroblocking and feel. So I can get very close results by increasing compression. So if there is only slight movement then a higher compression can be used but fast movement at low frame rates will need much higher values. Generally speaking I think you could get away with about 1/3 more media but many times come very close as if shooting on the 6k.

As James notes, RedCode compression does not scale in a linear manner. Short version: it's more efficient as the initial capture format gets bigger. The interplay between supersampling ratio and compression ratio is also a factor - with different strategies indicated for different pipelines. For clients who desire a more Alexa like workflow, scaling to 1080/2K or 2160/UHD/4K ProRes in log format should work nicely.

Furthermore, I want to echo James' notes on managing compression ratios with an eye to what you are shooting. 10:1 for talking heads, 5:1 for gimbal work, etc.

Cheers - #19

 

[Brendan H. Banks]

Why compromise from the beginning? Hard drives are cheap, get some 10TB ones and shoot at minimum compression. You can always lessen the quality of the image in post but you can't get it back if you start at a compromised position.

Well, when you want to shoot and the client or post team wants something specific, you make it happen. The other thing is doing commercial work vs feature work. I'm lining up a feature right now and shooting 7k will be an absolute monster for post and they're not sure they want it. 4k ProRes would mimic an Alexa workflow that so many post houses use.

At the moment, I'm working on upgrading my in-house post system to offer shooting + color services on the same project. They can edit in proxies and I'll do the online myself. It's a great idea in theory but the producers signing the checks eventually make the calls on these things.

 

[Patrick Tresch]

As James notes, RedCode compression does not scale in a linear manner. Short version: it's more efficient as the initial capture format gets bigger. The interplay between supersampling ratio and compression ratio is also a factor - with different strategies indicated for different pipelines. For clients who desire a more Alexa like workflow, scaling to 1080/2K or 2160/UHD/4K ProRes in log format should work nicely.

Furthermore, I want to echo James' notes on managing compression ratios with an eye to what you are shooting. 10:1 for talking heads, 5:1 for gimbal work, etc.

Cheers - #19

Hello Blair,

I was also thinking wavelet compression would have make some difference in the X2 factor between 6k and 8k R3D but Phil's chart told me other wise :


Weapon 8K Helium S35
http://phfx.com/tools/REDCamAssist/...solution=8K&format=FF&ana=none&fps=24&HDRx=no


Weapon 6K Dragon S35
http://phfx.com/tools/REDCamAssist/...solution=6K&format=FF&ana=none&fps=24&HDRx=no

Into two tabs. You can compare the REDCODE RAW Data Rates there.

The slight difference it that Helium FF is a 1.90:1 frame and the Dragon FF is a 1.94:1 frame.

Pat

 

[Patrick Tresch]

4k ProRes would mimic an Alexa workflow that so many post houses use.

Yes and coming from an 8k sensor instead of a 3.2k sensor will make it 4k for real!

 

[Blair S. Paulsen]

Whatever Phil's charts show is an excellent guide. That said, the impact of any R3D compression above 2.5:1 is a combination of mathematical and perceptual factors. We could start, or join, a whole other thread on that issue - but I think it's been beaten to a fine pulp already ;-)

I would also point out that the data rate one works at is not just about the size of the raster - 5K, 7K, 8K, etc - it's also about what compression ratio you choose to shoot at and what the maximum write speed is of your camera/mag. Like a lot of folks, I choose to keep the compression low and the frame large - data load be damned.

As has been pointed out, max data rate is not viable for every project. If you have a situation where any number above 4K is causing big pushback, then scaling from 7K or 8K to 2160/UHD/4K ProRes gives you a pretty good answer. If you expose well and use log encoded values in your ProRes files there should be plenty of meat for the colorist. If you do have a couple of scenes with lots of movement or low lighting levels in the mix, you might want to seek permission to capture those as RAW R3Ds and custom convert them with some extra finesse into ProRes.

Cheers - #19

 

[Mikael Lubtchansky]

The thing is as R3D works in wavelet based compression. If you take one of the available Helium 8k R3D's or a native 30mp plus photo and compress both at different JPEG2000 values you can easily reduce the 8k frame to the same apparent resolution and file size in terms of macroblocking and feel. So I can get very close results by increasing compression. So if there is only slight movement then a higher compression can be used but fast movement at low frame rates will need much higher values. Generally speaking I think you could get away with about 1/3 more media but many times come very close as if shooting on the 6k.

Furthermore, I want to echo James' notes on managing compression ratios with an eye to what you are shooting. 10:1 for talking heads, 5:1 for gimbal work, etc.

FWIW please note that the wavelet / REDCode compression is applied to the RAW, bayer image... it is not the same as downsizing an RGB image using JPEG2000... altho it is also based on wavelet.
You can expect much better results / efficiency when compression a monochrome image pre debayer...

Also REDCode is intra frame based compression not inter frame so any movement does not really make a difference... fine details are what will suffer most in a highly compressed image, be it a moving or static shot...