How RAW is Red RAW? (R3D)

[Graeme Nattress]

I guess you'll hear if we say we've improved compression or something that would need a re-shoot for tests.

Graeme

 

[Deanan]

Just go see CHE#1, there were a few spots that looked like compression

Che was recorded with REDcode 24 with what was basically an alpha version of the codec. There have been hundreds of improvements to the codec's performance, efficiency and quality since.

 

[Wesley Scoggins]

And they also weren't working with the original R3D's were they? They outputted to 2K DPX's like as early as possible IIRC?

 

[Deanan]

And they also weren't working with the original R3D's were they? They outputted to 2K DPX's like as early as possible IIRC?

They worked with the R3D up through the first DI pass.

 

[Sidney L. Plaut]

I think Che looked great and the argentine too!

only thing with Che was the scene in the apartment in the beginning it was very "grainy" - I just assumed they had underexposed it and brought it up in post?

 

[Deanan]

I think Che looked great and the argentine too!

only thing with Che was the scene in the apartment in the beginning it was very "grainy" - I just assumed they had underexposed it and brought it up in post?

That scene was shot on 16mm film as were the B/W scenes.

 

[Jonathan Smiles]

I recently compared some blue/green screen tests we shot in 2007 on Build 11 4K REDCODE 28 to the blue/green screen tests we shot recently on M-X Build 30 4.5K RECODE 42.

Big difference, so I would say you need to reshoot your tests with each new firmware to be sure.

 

[Ryan E. Walters]

I know I need to with a new sensor, as it is the same as shooting with a new camera. However, I was more interested in knowing about each new build and if I should re-shoot or just re-process. And it seems that the answer is that I need to reprocess most of the time, and reshoot only when major changes happen.

 

[Bob Gruen]

Uncompressed, REDOne, is 4096x2304x12bits (bayer pattern, remember so no x3)

Graeme

Graeme,

I was under the impression that the Red Cameras digitize the analog signal coming off of the sensor, in which case the math above wouldn't be appropriate. Doesn't the Red Team refer to processing the R3D files as de-bayering? This would indicate that the Bayer signal (an analog signal) is still intact in the R3D files. Do I have this wrong?

Bob

 

[Graeme Nattress]

There is no such thing as a Bayer signal - the Bayer refers to the pattern of the colours on the photosites of the sensor. The sensor, being CMOS, does A-to-D on chip, hence no analogue signal is ever output - it's all digital from the sensor outwards.

Graeme

 

[Jonathan Smiles]

This would indicate that the Bayer signal (an analog signal) is still intact in the R3D files. Do I have this wrong?

Yes, Bayer just means that the sensor has Red, Green and Blue filters arranged in a bayer pattern.

 

[Andrew M.]

Bayer sensor is one channel sensor. So no R or G or B direct acquisition of each channel like on some other cameras is done.

When you record from Bayer type sensors you read one and every pixel of the sensor at a time with whatever AtoD resolution is there. In case of RED it is 12 bit AtoD converter. This gives you 12 bits or 1.5 Bytes per pixel.
Now the storage capacity we usually express in Bytes so 8 bits.
Speed of the recording we express in bits/sec or more often now Bytes/sec.

So take the frame size that you record say 4096X2304 (it is how many pixels you have on the Bayer sensor for this given format) and multiply by 12 bits.
If you are shooting in 4K 2:1 (4096x2048) than you have to multiply these numbers.
Remember, RED is 4.5K 16:9 sensor so you can go much higher with number of pixels per frame almost all the way to 12MP.

So to answer your question, pick the format and then multiply horizontal times vertical times 12 bits and multiply times number of frames per second. This way you get the speed of completely RAW data as well as storage needed if you would capture completely RAW data. (not the case if you do not have the RAW port on RED camera)
Debayering process is not a compression process. It just takes the single channel Bayer stream and converts it to final format that usually is RGB format. Without compression, this process will create 3 channels, R,G,B with the same data size for each channel, so 3 times bigger.

It is not what RED is doing. RED is taking a single channel Bayer stream and compressing it about 12 times, producing single channel compressed RAW called REDRAW. One could compress this data more than 12 times but you will start losing details and quality of the 4K image that is the ultimate target.

Until recently compressing 5 times was considered virtually lossless. With the new science in this field this number will be higher and higher in time.

 

[severin schultze]

Graeme,

does the ADC in the R1 happen on the chip? Though it's an CMOS it generally doesn't need to happen on there?!

- Severin

 

[Graeme Nattress]

You're asking the wrong guy. I don't even know enough to play a sensor guru on television. It happens somewhere, that's for sure.

Graeme

 

[Gunleik Groven]

You're asking the wrong guy. I don't even know enough to play a sensor guru on television. It happens somewhere, that's for sure.

Graeme

I could try to get you casted... -