From REDCODE RAW to DNG?

[Trevor Meier]

Any thoughts on allowing export from RedCine to DNG files? This would allow taking RED footage into a tool like Lightroom or Photoshop for a custom look. REDCODE RAW to DNG should be a relatively simple format translation... no debayering or other processing required.

Reason I'm asking... I'm unhappy with the colour fidelity and manipulation tools that are currently available for digital video until you get in the high end. A Baselight allows for fine control of images at very high quality.. but there's no way to get close to that image fidelity for less than a few hundred thousand.

However, in my photography business, I've been used to having that sort of fine-level image control for years. The tools already exist for a few hundred dollars. They're not designed for moving images, but there are methods for dealing with that problem.

We're getting there... it'll be interesting to see what Apple does with Final Touch, for example... but RED is about creating options. And... well, there's not a whole lot out there within indie reach for onlining 4K moving images just yet.

[Rob Lohman]

Do you really need that? The only thing that you should be able to do extra is use a different debayer algorithm. All other controls should be available for regular RGB data.

Just export unprocessed 16-bit RGB stills from REDCINE and load those up into Photoshop or any other package you want to use.

[David Newman]

Rob,

I think DNG support would be nice. If you want to perform the same types of color information DNG offers on a 16-bit RGB file, for a start the RGB file is 3 times bigger, many times slower and typically RGB is gamma corrected whereas DNG is linear. It is not possible to offer every color correction feature that Lightroom or Photoshop offers on DNG within REDCINE, yet DNG exports would be much simpler than any other file format you intend to support. Without the metadata the DNG export will take you a couples of hours (or less) to develop -- it is a fixed header size (with limited issues, just copy a header off another DNG to start with) with 16-bit RAW linear data straight from the sensor.

[Matthew Greene]

Hi David, if you don't mind me asking I'm curious as to where CineForm sees itself in regards to RED and it's users.

[Emanuel A.]

David,

Since we have 10-bit log, 10-bit linear or 12-bit linear as full_options@REDCODE, your 12-bit skepticism is not a problem anymore or don't you agree?

[David Newman]

Hi David, if you don't mind me asking I'm curious as to where CineForm sees itself in regards to RED and it's users.

We will support the users with our Prospect 2K/4K products as much as RED allows, and probably a tad more (consider that Panasonic wasn't supportive in CineForm adding HVX200 support -- we did it anyway.) We know REDCINE will output to QuickTime based compressors, so CineForm Intermediate will work fine. For CineForm RAW support we have requested unmodified (no-demosaic, no color processing) RAW data from REDCINE, and RED is thinking about it. Also we would love to offer an external recorder to support 4K at 60p and 2k at 120p. Email RED or post here to encourage a open design for third parties.


Since we have 10-bit log, 10-bit linear or 12-bit linear as full_options@REDCODE, your 12-bit skepticism is not a problem anymore or don't you agree?

Absolutely agree that 10-bit log (depends on the curve) will be the highest quality mode for any lossy image compression (particular those compressed much more than 5:1, which REDCODE will be.) 10-bit linear is an odd offering as even 8-bit 709 with a 2.2 gamma is superior (8-bit 709 has more shadow precision than 10-bit linear uncompressed.) Linear compression is cool but only for mathematically lossless compression applications, or uncompressed.

[Trevor Meier]

Just export unprocessed 16-bit RGB stills from REDCINE and load those up into Photoshop or any other package you want to use.

Unprocessed 16-bit files would require almost 4x the storage as RAW 12-bit, and require far more processing time. As well, whoever's running RedCine will have to take care to preserve highlight & shadow detail since it's being baked in.

[Roberto B]

Absolutely agree that 10-bit log (depends on the curve) will be the highest quality mode for any lossy image compression (particular those compressed much more than 5:1, which REDCODE will be.) 10-bit linear is an odd offering as even 8-bit 709 with a 2.2 gamma is superior (8-bit 709 has more shadow precision than 10-bit linear uncompressed.) Linear compression is cool but only for mathematically lossless compression applications, or uncompressed.

and the 12-bit?.. linear, of course.. because log there isn't there..

[David Newman]

and the 12-bit?.. linear, of course.. because log there isn't there..

Yes, 12-bit log has value, although it is not intintive to use it from a 12-bit linear sensor; understanding compression and the human visiual factors make it worth while. 12-bit log should be added, that is what we are are adding to CineForm RAW.

[Rob Lohman]

Unprocessed 16-bit files would require almost 4x the storage as RAW 12-bit, and require far more processing time. As well, whoever's running RedCine will have to take care to preserve highlight & shadow detail since it's being baked in.

You could disable all processing and leave the image linear. Basically just the debayer has been applied.

Yes, the files will be larger. But the processing time will not if you have a fast enough RAID. The target applications converts the RAW to RGB anyway. So the only downside is more disk space and perhaps the loss of some RAW tools in Photoshop (or other DNG supporting applications).

I'm not saying this won't be in it, but it isn't at the moment. DNG files are a special form of TIFF files so not that hard to generate. It however does take time to develop and more importantly to test. We have to weigh what kind of features to get into this first release.

[Roberto B]

Yes, 12-bit log has value, although it is not intintive to use it from a 12-bit linear sensor; understanding compression and the human visiual factors make it worth while. 12-bit log should be added, that is what we are are adding to CineForm RAW.it means will you provide the 12-bit log for Red Code users from Red Cine?.. sorry me but all these log and linear, can anyone explain what do all these new words (for us) mean?..

edit
btw, the Red specs are referring only 12-bit linear not 12-bit log!.. we (the common guys) presume there will be some difference..

[David Newman]

Sorry for the confusion, discussing several things at once. CineForm will be offering high-performance post-production solutions in some form for Red users, we hope Red will allow for tighter integration. The linear vs log argument is separate issue that can get complicated. Linear is simple, if the pixel value is twice as bright, it is stored with a value twice as big. Log applies a curve so that a doubling of brightness results in a smaller increase in store value. In the end both can store the same image data. The issue is the effect of compression, which adds subtle distortion. This distortion is visually more apparent in the shadows when using linear compression (which is bad for your shadow detail), yet a correctly selected log curve greatly limits shadow distortion and hides the distortion in a more visually acceptable manner.

[Roberto B]

thx for the explanation..

so, 12-bit log is better..

we presume Red Codec will be 12-bit linear and 10-bit linear as well 10-bit log..

which will it be the best?.. 12-bit linear (giving 4096 levels) or 10-bit log with such properties (but only w/ 1048 levels)?

if both are good for different purposes, can we mix both according our different needs?.. or there won't be because log is better than linear and period

edit
and will it be possible to go to the 12-bit log from Red Cine?.. including your 4k for $5k?.. as far as the open design goes, it seems the Red Cine will have open doors so..

[David Newman]

For almost all imagine senarios 10-bit log (using a good curve) will out perform 12-bit linear (at the same bit-rate.) The one exception is when you need to invert the image with a negative effect, but that is not a popular look.

There little value in doing 12-bit linear to 12-log in RedCine as the shadow data would already have been lost by the camera's compression. The curves need to be applied before the camera's compression stage.

I hope you are correct about "open doors", only time will tell.

[Emanuel A.]

Thanks David! It is so much appreciated your help. You've been opening your knowledge doors and this is a value-added that we future RED buyers can't forget. I believe if your contribution will mean a better offer, some of us will be crossing your borders. For sure, there is a law that rules this relationship. There will be a balance between the demand vs. offer (where all the players have their chance to show what you competitive providers can supply) that will give all the answers. But you have all the available technology you know to handle with it. And this competition will be a health gift to all RED users. Including RED team. I'm a liberal believer...it's like maths: a few figures can fit (even altogether sometimes :D) but there is one which fits the best!


For almost all imagine senarios 10-bit log (using a good curve) will out perform 12-bit linear (at the same bit-rate.)So, since we don't have 12-bit log @REDCODE, will the 10-bit log be the best REDCODE shooting route?


There little value in doing 12-bit linear to 12-log in RedCine as the shadow data would already have been lost by the camera's compression. The curves need to be applied before the camera's compression stage.Understood.

Apart your external recorder solution (for slow motion purposes the variable frame rates will be helpful such as 4k@60fps or 2k@120fps which can be useful, that is, if RED will allow the downsampling via from the sensor in this latter case -- that I'm not sure if it will be possible or am I missing something over this my equation?), then doing 12-bit log in REDCINE from 12-bit linear is useless since we're coming from the 10-bit log (REDCODE as well).

But the REDCODE 12-bit linear can be useful when there is no shadows to deal with and the lack of contrast (or its neutrality) can give some good results into the 12-bit linear realm.

Am I right in my thoughts?

[David Newman]

But the REDCODE 12-bit linear can be useful when there is no shadows to deal with and the lack of contrast (or its neutrality) can give some good results into the 12-bit linear realm.

Am I right in my thoughts?

It is a odd image to be optimizing for, something that has no shadows. Even so I still favor the 10-bit log approach for all partical image shooting conditions. The reason is in how the human eye perceives, I believe you can hide compression distortion (artifacts) easier using a log curve than using linear coding. There will always be images that might break the general rule, and as bit-rates get higher the 12-bit vs 10 will eventual exceed the advantages of log vs linear, I just don't believe 27MB/s is high enough to go linear for 4K RAW. I've been working with wavelets for nearly a decade, so some of this is just instinct now.