Uncompressed vs REDCODE

[stephan orlandic]

People here think that the Milk Girls footage was uncompressed. It was REDCODE. There were many things along the way that changed in a RED ONE that caused image issues, but REDCODE was not one of them. REDCODE had to change to compensate for other things in the chain."

1.
the Milk girls clips were only available as recompressed quicktime hd.
And there were no red raw files, or uncompressed image sequence so they could be properly analysed.

Most people probably watched it on some computer lcd or notebook,
again, in recompressed hd quicktime, which can not by any means,
be a reference to make a delicate judgment like the above mentioned.

2.
A great deal of redusers at the moment of watching milkgirls had very limited or no experience with high resolution digital cinema files, or high resolutions scans.
Therefore, their observation of limited quality clips,
should not be a reference to RED company, and mr. Jannard.
Espacially, it should not be used as a statment that RECODE is almost perfect,
and that noone can say that it is compressed.

is anyone really unhappy with REDCODE? I haven't heard that one

Peter Jackson couldn't tell them apart. Actually, no one has been able to.

Let me start with this,
it is very obvious for a qualified person (CTO, colorist, post or vfx supervisor, equipment purchasing consultant, ...) to see an image quality differences between different film scanners (Northlight, ARRISCANNER, Ditto, Director, Spirit) which all provide uncompressed 2k or 4k files 10bit DPX or 16bit TIFF files.

It is impossible not to see a difference between uncompressed RED RAW,
and the RECODE 36 in technically well prepared conditions.

Since we don't know anything how the screening was setup ( resolution, type of files, projector, server, vcr, ...) , I can't make any claims what
went wrong.

However, it could be possible that machine-server which reproduced both materials was not properly setup, so all the beauty of RED RAW couldn't
come up to the screen. Or the RAW files were not prepared as they should be. It could only be a guess since there are no details about screening.

"Peter Jackson and his team couldn't tell the difference with a RED ONE, in fact... guessed wrong."

At the end, it would be a great shock that even by worst conditions,
Peter Doyle or Andy Cole wouldn't see a difference.
Of course, it they were there.


For anyone interested I could upload many examples of compression issues, starting from Monday.


No disrespect or offence meant.


Best Regards

Stephan

 

[Fredrik Callinggard]

I just want to take a moment and say that no one is saying that RED Code isn't beautiful. Shoot right you can fool anyone from believing it was shot on film.

That's not the problem. The problem occurs when you have to dig deeper into the image. We all know that the blue channel is particular weak and that's IMO the general issue.

I think this is a great article made by Art Adams at Provideo Coalition - it explains the issues together with some great illustrations :biggrin:.

http://provideocoalition.com/index.php/aadams/story/red_and_green_screen/

 

[GlennChan]

Perhaps it's not the compression that is the issue here... unless you hit a particular build where there are codec bugs. There are other codecs out there that do a lot more damage and are still keyable (blur and erode the crap out of that matte you know?).

Have you considered:
1- Perhaps the way the image was exposed and processed in RA/Redcine caused noise to go up. Take a look into how DRX works (compare between 1 and 0), use the histogram warnings to not clip any highlights, etc.
2- Tungsten lighting can cause a very noisy image. Watch out!

I think the compression bashing here is misplaced. An example would be the show Lazytown... for workflow reasons, then no longer online from DPXes but from DNxHD files (according to the Avid website anyways). And that show has probably made some crazy amount of money.
They've done a great job with keying (from what I can tell from watching the show on cable). The main nitpick I see with their job is on motion blur... more sensitivity + higher shutter would be helpful. I don't think compression is a problem for them as they are adding it into their workflow. (And it is highly likely that Redcode is way better than theDNxHD codec, which is older technology.)

 

[Eric Lange]

Dear RED team,

I think it's all fine of Jonas to bring this up. He's not suggesting that you should consider uncompressed. He's merely saying that RED has a great market in his business (one of Scandinavia's most prestigious and biggest post houses). That your cameras are involved in Scandinavia's high end market, but is now declining slightly because of this.

I've encountered these compression problems many times and is learning how to work around it, but knows that many productions will not take the risk after being burnt ones. With me there's no problem since they know my background and can see many successful RED commercials on my reel.

BUT if I was to shoot for example green screen and it's a little more elaborate in it's lighting then straight forward (still talking about well lit green but around it). Well then I wouldn't shoot RED Code because of it's compression. I've had a 2 mayor green screen shoots were we had a lot of "cleaning" up to do.

(PS I now know how to shoot RED Code green screen successfully but it would undermine my creativity).

So I agree with Jonas that just like me seems to like your camera and the look of it (I've actually grown into liking it more then film). That just bare this in mind when you're working on it for the future.

The 2 achilles heels right now, IMHO, are rolling shutter issues (strobe and skew) and compression when it comes to post production. Strobe and compression being the biggest thorns in my side.


regards,

Fredrik Callinggard

I think its very important (of course ) to understand the limitations of a particular system so that you can get the most out of it or be able to put your best foot forward. We are just contemplating/developing some chromakey capability for a new software that very cleanly extracts new and various types of texture mapped models for 3d composit/vfx work from stereo live action footage (S-3d). We are mainly software engineers and technical photographers/photogrammetrists.

So, on a practical note could anyone kindly recommend the optimal conditions and techniques to get the best out of green screen with RED one? [I read the link offered above from provideocoalition] In essence we are not necessarily hampered by “creative freedom” but rather we are looking to get the best technical test results for our software that may also end up as tutorial data sets when the software ships.

Any help or advice beyond (don’t shoot unfiltered tungsten) would of course be greatly appreciated?

(What also would be the practical differences in using a D21 instead?) [We are keen on RED as we like higher res imagery and large sensor size for our type of work]

Cheers

Eric

 

[REDneck]

I think its very important (of course ) to understand the limitations of a particular system so that you can get the most out of it or be able to put your best foot forward. We are just contemplating/developing some chromakey capability for a new software that very cleanly extracts new and various types of texture mapped models for 3d composit/vfx work from stereo live action footage (S-3d). We are mainly software engineers and technical photographers/photogrammetrists.

So, on a practical note could anyone kindly recommend the optimal conditions and techniques to get the best out of green screen with RED one? [I read the link offered above from provideocoalition] In essence we are not necessarily hampered by “creative freedom” but rather we are looking to get the best technical test results for our software that may also end up as tutorial data sets when the software ships.

Any help or advice beyond (don’t shoot unfiltered tungsten) would of course be greatly appreciated?

(What also would be the practical differences in using a D21 instead?) [We are keen on RED as we like higher res imagery and large sensor size for our type of work]

Cheers

Eric

I don't know if it's always this way, but I've seen HD dailies of two movies shot on the D21 and both were surprisingly noisy. Like, very noisy.

 

[Dylan Reeve]

So, on a practical note could anyone kindly recommend the optimal conditions and techniques to get the best out of green screen with RED one? [I read the link offered above from provideocoalition] In essence we are not necessarily hampered by “creative freedom” but rather we are looking to get the best technical test results for our software that may also end up as tutorial data sets when the software ships.

There was a link to a ProVideoCoalition article a few posts back that seems very helpful. Can't say I've tried what they suggest, but they seem to have tested well.

I've been involved with a number of chromakey RED shoots now and have generally found th results very good. Even when suffering from tungsten lighting we found the key was still very recoverable with some tweaked processing for the RED RAW file.

Certainly the REDCODE codec is compressed, and in certain circumstances that compression will be apparent, but so far in the work I've been involved with it hasn't become a problem. All formats have their issues, RED is no exception.

 

[Fredrik Callinggard]

Certainly the REDCODE codec is compressed, and in certain circumstances that compression will be apparent, but so far in the work I've been involved with it hasn't become a problem. All formats have their issues, RED is no exception.

Absolutely but film still stands better if your in a scenario we're for example you have a lot of low light levels on your objects (shadows or dark material) and/or there's a lot of reflective/ see through material.

All of them being harsh problems to the blue channel.

Main problem for me has been that the compression noise is a compression artifact. Redcode has some kind of "streak" in it, which is very difficult to deal with.

I had a job were there's normally maybe 3 nodes to clean up (when on film) to make a key. Well on our RED material we had to use 18 nodes to be able to clean it up. All channels together looked stunning, but blue channel singled out, broke. There was a lot of shadows and reflective material that was affected by this and needless to say we spent a month in only cleaning up.

If I had a straight forward green screen shoot which was completely evenly lit in every way, with daylight, I would not hesitate. I'd shoot on RED. But as I said it does limit you creativity.

 

[Petr Dvorak]

If you disagree I like to post some stills where you can really see what I'm talking about.

Jonas Drehn
Duckling A/S
www.duckling.dk

Jonas, 1 image is more than 1000 words.
So don't hesistate a second and post stills and 1:1 crops with appropriate explanation for us less experienced ones.
I am hungry to learn more about it.
Thank you.

Devil is in details.

 

[Eric Lange]

Jonas, 1 image is more than 1000 words.
So don't hesistate a second and post stills and 1:1 crops with appropriate explanation for us less experienced ones.
I am hungry to learn more about it.
Thank you.

Devil is in details.

I second that, would be (very) nice to see specific artifacts and also their “cure” as far as is possible.

Some of the posts seem to be a bit “down” on the blue channel from RED ONE. I used to work with a variety of very high end film stocks, neg and diapositives used in aerial photography/reconnaissance and these were scanned on some the most expensive scanners on the planet (about ten years ago) and with these “films” you would still get a weak or noisy blue channel. (Have also found the same to be true when shooting close range even on subjects that were pre-dominantly blue.)


Is the “blue channel” on scanned film really that much better than the blue channel on digital/RED One or a D21? (I have no direct experience of this)

thanks for the posts so far on chromakey technique and RED,

Cheers ,

Eric

 

[M Most]

I don't know if it's always this way, but I've seen HD dailies of two movies shot on the D21 and both were surprisingly noisy. Like, very noisy.

It's not always that way. Never judge a camera based on footage from one DP or one production. Not D21, not Red, not anything, including film stock. Especially when the DP is not you.

 

[REDneck]

It's not always that way. Never judge a camera based on footage from one DP or one production. Not D21, not Red, not anything, including film stock. Especially when the DP is not you.

two productions. two DPs. same camera.

 

[M Most]

two productions. two DPs. same camera.

What are you using to convert the RAW files? And with what settings?

 

[REDneck]

i'm not working on either show, but they captured to HDCam SR and ingested with Prores 422 HQ. from quick looks at several scenes, I got the impression that it is noisy. maybe the DPs don't know how to work it yet?

 

[Fredrik Callinggard]

Some of the posts seem to be a bit “down” on the blue channel from RED ONE. I used to work with a variety of very high end film stocks, neg and diapositives used in aerial photography/reconnaissance and these were scanned on some the most expensive scanners on the planet (about ten years ago) and with these “films” you would still get a weak or noisy blue channel.

It's not noise that's the problem with redcode, it's the compression. It's the artifacts that is perceived as the noise which is the problem. And yes it's always the blue channel which is weak on all cameras, but since it's a bayer sensor in the RED, the blue channel is even more so.

 

[Jarred Land]

It is impossible not to see a difference between uncompressed RED RAW,
and the RECODE 36 in technically well prepared conditions.

Since we don't know anything how the screening was setup ( resolution, type of files, projector, server, vcr, ...) , I can't make any claims what
went wrong.

Hey Stepen no disrespect, but The test Jim is referring to was not at RED.. it was in one of the 4k reference facilities at one of the biggest studios.. that was completely calibrated the day before the test... And it was not just Peter there.. it was a whole bunch of very qualified individuals who's job it is to tell the difference.

We all know REDCODE isn't perfect, we make it better all the time. But some of the artifacts that you may be seeing may not be caused from REDCODE itself, but may be from a poor implementation of REDCODE (that we indeed very well have done in some of the builds in the past), or possibly from user error.

( i.e i have seen alot of really crappy H.264 encoded material, and some really good encoded H.264 material, its easy to blame the codec for user error, or lack of knowledge )

 

[Blair S. Paulsen]

RedCode 250 anyone?

RedCode 250 anyone?

I have worked with every build since the release of the camera in a wide variety of lighting situations. IMHO the most telling changes have often been dot releases that appeared to dial back the image quality a hair in response to codec errors. This is pure conjecture as to WHY the level of compression artifacting has varied, but unless I need a trip to the eye doctor it is obvious that the short history of RedCode is one of twists and turns.

That said, the good news is that milder compression is on the menu for the new rigs. How awesome would it be to be able to choose your compression rate based on project requirements and/or recording media capacity? Sign me up!

I can't wait to see what Graeme and the Red Team can do with a bigger bit budget. I am imagining a 5K Bayer file compressed in camera to around 140MBytes/sec, the first off camera processing step would include demosaicing and downsampling to a 2K (or 1080) 10 bit RGB file for offline and a 4K 16 bit RGB file for grading/conform.

 

[Anders Holck]

Jonas brings up an important issue.

It's not really the noise in the blue channel, but the point where the organic camera noise dives into compression noise. No need to defend the codec and say it isn't there, because it indeed is. Whether you can live with this artifact is of course a matter of taste.

Shooting 4k, daylight with a rich exposure and you might not experience the problem, but this can still pop up occasionally if you utilize the bottom stops of the dynamic range.

Photographic style is also a determining factor in this, because some WILL prefer tungsten lighting, even without the lighting budget to go 160ASA with a 80C filter, and some will combine that with a slight underexposure to protect the highlights.

And then combine that with 2k resolution for high speed, and redspace viewing and you will experience problems. This is not an operator error. And we still see this in build 17 using the newest tools.

It's important to realize that Jonas isn't saying that the camera doesn't deliver beautiful images because it does and his post house has embraced the RED from the get go (He even had a reservation at some point).

But what he is saying is that his clients are beginning to look for other options because of this issue, which will be a problem for all of us.

This is a balancing act of praising the good aspects, but still be serious and inform about the downsides, everything else would simply be unprofessional. But it is of course also important not to blow this out of proportions, because redcode does perform very well most of the time.

 

[Eric Lange]

This should be called the brotherhood of the anti aliased pixel, where for some applications you literally have to split hairs!

So what some of you are saying is that its difficult to compress (high frequency) speckly noise from blue channel or dark current in a spatially accurate way and efficiently without compromising the whole efficiency of the compression scheme; (probably not a big surprise) I.e. theoretically one could effect accurate representations, but you wouldn’t be able to shoot at 24fps (ish). So in terms of various bandwidth trade off s, this is the most sensible thing to do as such artifacts are normally invisible for a “regular” RGB image thrown up on the screen. I.e. it is computationally very inefficient to accurately represent high frequency noise in any compression scheme.

So I think I seem to understand some of Jim’s frustration about requests for RAW or near raw, as it’s the “on the fly” compression as the scan lines are being acquired that makes that RED ONE function ( I am guessing) so that you can shoot at 30 or 60 fps. Presumably with the line of new cameras, it’s the gargantuan bandwidth that will allow a 3:1 compression ratio, so perhaps by the end of the year these types of problems may be done and dusted.

We/(my company) very much have to consider what is happening at the pixel scale as we try to employ and develop various 3d extraction algorithms from stereo footage and stills. Standard algorithms are not very intelligent and tend to be fooled or trip up on compressed imagery as the algorithms try to look for unique matches to extract 3d. When the images are composed of repeating wavelet blocks you tend to get a lot false matches and hence spurious 3d points. The issue for us is do we spend development time and funds on developing pattern matching algorithms that are not fooled by compression artifacts or do we wait for 3:1 compression that may be spatially more lossless? I.e. you spend a lot of effort to solve a problem and by the time you finally get there the problem has vanished and no longer needs to be solved!

 

[conrad gaunt]

Camera 1912.. produces amazing 2k/1080p downsamples, from the 4k bayer, compression artifacts aren't an issue.

Now, unfortunately, 2k doesn't allow this level of downsample, and viewing 4k at 1:1, isn't so pleasing either IMHO

The R1 2k is best for generating 720p downsamples.

 

[M Most]

i'm not working on either show, but they captured to HDCam SR and ingested with Prores 422 HQ. from quick looks at several scenes, I got the impression that it is noisy. maybe the DPs don't know how to work it yet?

If they captured to HDCam SR, they weren't capturing RAW data. They were capturing over HD-SDI, which is a processed signal. This was further complicated (at least in the noise department) by compressing using ProRes. My experience with the D21 is based on the RAW files, delivered over T-Link to an S.two recorder, and converted to RGB via Arri's Image Booster software. The material I've dealt with was shot anamorphic, and is very, very clean, very sharp, and in general, very nice.