Benifits of Uncompressed 10bit vs DPX vs Tiff vs Cineon vs ProRes 422 HQ

[BigLu]

I am very interested and would like to understand the benefits and most logical usages for all the top codecs.
Many argue and say use this use that.
Some swear by ProRes 422HQ but when I send files to high end post effects houses they reject it. They want it sometimes in Uncompressed 10bit accept 8bit or want DPX. Ive even had to send out all TIFFs on a music video.

So I want to hear from someone of you who deal with high end codecs.

DPX vs Uncompressed 10Bit 4:2:2 is the one i want to start with and really understand.
But please share your thoughts on the other codecs


DPX
TIFF
Cineon
Uncompressed 10bit 4:2:2
Uncompressed 8bit 4:2:2
Do you prefer AJA's codec of uncompressed or Blackmagics
and
ProRes 422 HQ

 

[Stephen Williams]

For high end work DPX would be my choice, TIFF is for post houses who have never worked with film or are frightened of getting Log images that need a LUT! TIFF images are only 8 or 16 bit, usually they are used to working with 8 bit I guess, if you gave them 16 bit it would be OK but huge files.

 

[Kaku Ito]

Luis,

When you compare from DPX to video format, the video format that can get close to DPX is 4:4:4 format, not 4:2:2.

In the broadcast centered world, since that is 4:2:2 centric, working with 4:2:2 is appropriate, but when you go high-end for film work, the least you should stay for RED footage is 4:4:4.

 

[Shawn Booth]

Aloha Lu -

UC10 bit. Preferred codec whenever possible.

DPX. (Professional use, visual fx work, finer control of your image)

TIFF - lower end budget/pro, down/quick 'n dirty work.

ProRes - I use this for the low budget shows that don't provide their own storage.

My .02

 

[BigLu]

Sweetness
So it looks like im leaning towards UC 10 bit
now do you guys have a preferance between AJA's codec or Blackmagics codec
or is that specific to what card i buy?
Does it matter?

 

[CraigDavidson]

DPX is a file format, and says nothing of the contents. Like a QuickTime movie, just about anything can be placed in a DPX file. Some 4K cameras record RAW uncompressed to DPX files. The format also supports YUV and RGB formats of various sizes. For RED cameras, a DPX is normally 10-bits for each of the RGB values stored in a 32-bit word. The header contains the information about the pixel format, curves, LUTs and everything else useful in reproducing the image.

YUV is traditionally thought of as a 422 format. For each luminance value, Y, you get either a chroma or a tint. This makes for a smaller file, but you do loose some detail.

Uncompressed 10 bit YUV is just that, uncompressed. The AJA or BlackMagic drivers difference would apply only for the display or output from the card. For file output, the critical path is how the YUV is created. The RED source is natively RGB, so there is a conversion involved.

A YUV file could be 444, a chroma and tint stored with each luma, but there is no benefit to do it this way over an RGB format, plus for the RED source you still need to do a conversion.

A number of post houses, such as FotoKem in LA, will accept CineForm files, either 422 or 444.

 

[M Most]

A number of post houses, such as FotoKem in LA, will accept CineForm files, either 422 or 444.

FotoKem is not a "high end post effects house." They are a lab and finishing facility, not a visual effects companies. In the world of visual effects, compression is unacceptable because it interferes with the ability to extract the cleanest mattes and manipulate the image.

 

[Dan Hudgins]

Certainty?

Certainty?

With uncompressed 48bit TIF at video gamma 2.2 there can be no doubt as to what the end result should look like, you hold more than the 12bits RED ONE puts out, you do not get double compression losses, and you get full resolutuion chroma. You also get more data for later adjustments than 30bpp 10bit DPX can hold, so histagram gaps do not develop.

If you only have 10 bits in a DPX the LUT/gamma adjustments in the film recorder can spread the brightness values so you end up with gaps in the D2A values going to the CRT/LCD/Laser, if you start with 16 bits per primary you can make large gamma adjustments in the film recorder before gaps develop.

 

[CraigDavidson]

With uncompressed 48bit TIF at video gamma 2.2 there can be no doubt as to what the end result should look like, you hold more than the 12bits RED ONE puts out, you do not get double compression losses, and you get full resolutuion chroma. You also get more data for later adjustments than DPX can hold, so histagram gaps do not develop.

There is nothing in the specification of DPX that restricts it to 10 bit RGB. DPX can hold 48bit RGB, or even RAW.

 

[Dan Hudgins]

Support?

Support?

DPX can hold 48bit RGB, or even RAW.

How common is it for graphics software to support 48bit RBG in DPX files for both read and write? Do you know of a list of programs that support that 48bpp DPX file type format?

How many support RAW DPX?

Support for 48bpp TIF is a bit problematic, but seems better than (48bpp) DPX support for freeware programs, maybe?

What advantage would there be to use 48bpp DPX over 48bpp TIF?

As far as RAW goes, DNG seems to be the target file type with wide support, Acam dII is going DNG, so who would use DPX RAW BAYER?

Edit: See this #15 post by Graeme Nattress,

http://reduser.net/forum/showthread.php?t=25979&page=2

It seems the RED-programs do not support more than 10bits for DPX.

 

[CraigDavidson]

How common is it for graphics software to support 48bit RBG in DPX files for both read and write? Do you know of a list of programs that support that 48bpp DPX file type format?

How many support RAW DPX?

Support for 48bpp TIF is a bit problematic, but seems better than (48bpp) DPX support for freeware programs, maybe?

What advantage would there be to use 48bpp DPX over 48bpp TIF?

As far as RAW goes, DNG seems to be the target file type with wide support, Acam dII is going DNG, so who would use DPX RAW BAYER?

Edit: See this #15 post by Graeme Nattress,

http://reduser.net/forum/showthread.php?t=25979&page=2

It seems the RED-programs do not support more than 10bits for DPX.

Raw DPX is the output from a Dalsa 4K camera.
DPX is designed as a container for image sequences, such as movies, and the header has fields that support the workflow. There may not be an advantage is basic image quality, but the workflow with the DPX should be easier.

The number of bits you need in the file depends on the use downstream. The camera records with 12 bits because the world is unfortunately very high contrast. As the image is processed, you need the extra bits to get the image tonality that you saw in the scene. The processing is done in floating point to avoid the introduction of rounding errors which would show up as banding. As Graeme said in the post you referenced, if the image is correctly exposed, "you will not loose going to 10bits for DPX"

Back to the original question, the preference for output to EFX would be 10bit 444 DPX files. As Kaku Ito said above, if it is film, you should avoid the 4:2:2 formats.

 

[Dan Hudgins]

Thanks...

Thanks...

Thanks for the info,

Raw DPX is the output from a Dalsa 4K camera.

Does Delsa or anyone have a converter to convert their Raw DPX frames to 48bpp TIF?

There may not be an advantage is basic image quality, but the workflow with the DPX should be easier.

Which "meta-data" filelds are you speaking of, SMPTE time code for the frame number?

The processing is done in floating point to avoid the introduction of rounding errors which would show up as banding.

I do not think anyone can see banding in images processed with 16bit LUT on 12bit data converted to 16bit data unless the tones are spread to the point that they are less than 9 or 10 bits and you would get just about as much banding with floating point at that contrast.

Floating point used on the pixel values is not required for "normal" grading that can be done with a 16bit LUT, can you explain more when it would be needed?

As Graeme said in the post you referenced, if the image is correctly exposed, "you will not loose going to 10bits for DPX"

He must have ment that when you have exposed so well and not too dark that the grading done after export is very slight or not at all so that you hold histogram gaps to give you 9bits or more of real data.

The limit of seeing banding depends on the dithering of the images, in low noise images you might see it a little at 10bits but most people will probably not, with noiser images you might start to see it at 9 bits. At 8 bits it can be seen in low noise images for out of focus areas of the images, some of the time. With very grainy film scans you might get away with 7 or 8 bits.

You want to have 10bits in the filmout, if you grade the images from 10bits you get histogram gaps unless you blur or downsize the images.