De-bayer Module

[J Davis]

Is this a feasible idea?
Would it be cheaper than an RED Rocket card?
My original idea was to use the brain but that asic is for FF1080p.

 

[albert rudnicki]

It would, but by doing it red would cut the sale of their RedRocket card.
I am sure both are very similar

 

[Gavin Greenwalt]

Is this a feasible idea?
Would it be cheaper than an RED Rocket™ card?
My original idea was to use the brain but that asic is for FF1080p.

So are you wanting 4k full debayer? I'm sure the brain is already doing that and then downsampling to 1080p.

 

[Dan Hudgins]

Half-size

Half-size

Its easer and faster to do 1/2 size or 1/3 size image and the results are not bad since the sensor is so large to start with, but what takes time is the chroma filters, OLPF sharpen, and many other steps that are required to get very good results.

Then on top of all the processing required you need to grade to compensate for shooting under natural light.

If you use a TV camera with its limited range under flat studio lights and use a wave form monitor you can go "one light" for use, but filmmaking under various kinds of light is the whole point of shooting "RAW" or rawish like REDCODE, so you can have enough range to grade later.

If you try to do the clipping in the camera then you are back in the same problems that TV cameras have and the many H.264 cameras, you need to light for what you get and live with it.

Although its nice to use the large sensors, going direct to clipped RGB results would have limited use because you would need to be dead on with the lighting since you drop 25% or more of the sensor data on the way to the RGB, with the RED programs you have DRX control to "recover" some of the lost data, but for "live" conversion you cannot go back and make DRX adjustments, since that data has been clipped off.

In effect REDCODE is a conpressed RGB format, just one that has not been clipped yet, so is like Linear DNG rather than Bayer DNG, except that the RED and BLUE images have less resolution than the GREEN image since the Bayer filter has fewer RED and BLUE pixels to start with.

If you want to speed up the results, then doing a half or third size "workprint" converion can be fast, with the de-Bayer program I am working on my "workprint" modes are maybe 10x as fast as full de-Bayer, but I would not recomend using the "workprint" output for anything serious because the point of shooting RAW is to get better results than a in-camera de-Bayer camera can because a PC can work in non-real time.

Converting from media to media is what you do on the PC now, so I guess you mean in real time, well REDROCKET in a PC is that function, and to put that in a module would be as large as a PC and use as much power about?

Its interesting that most of the buyers of the RED ONE do not seem to have been "filmmakers" in the feature film sense, and come more from a video background and so were looking for a video camera with "finished" compressed results rather than true RAW "digital negatives" that required days to process, like movie film. RED seems to have been walking a line between "filmmaker" type use and "videographer" type use and made something that has a foot in both worlds. Filmmaking will always be a smaller market than video since so much of video is for broadcast type use rather than the smaller theatrical market. Just for sales, you can wonder if they dropped the "filmmaker" aspects of the camera and just went for the "videographer" market would they in the end make more sales, like what must have been the judgments in Japan since they were ignoring the market RED has taken over?

I wonder what the ECN sales ratio was for Cinema use and for Broadcast and Industral use, I would think Cinema use in the last 10 years was much less than that used for TV commercials, does anyone know the specs of Kodaks sales to various markets of camera negative?

 

[Jannard]

Just be very careful that the workprint/offline at half-res doesn't somehow end up as the final print. You would be wasting a lot of energy to get inferior results. You would be surprised at how many are doing this... using half-res debayer for the whole pipeline. Ugh.

Jim

 

[Lee Saxon]

Just be very careful that the workprint/offline at half-res doesn't somehow end up as the final print. You would be wasting a lot of energy to get inferior results. You would be surprised at how many are doing this... using half-res debayer for the whole pipeline. Ugh.

Jim

Just stick a "THIS VERSION SUCKS" watermark on it

 

[J Davis]

So are you wanting 4k full debayer? I'm sure the brain is already doing that and then downsampling to 1080p.

Sure but its hardwired to sensor output. What I'm suggesting is something that is for through put processing as per diagram. Full res or half res debayer would be selectable. All the elements to make this happen are pretty much already there.

It would, but by doing it red would cut the sale of their RedRocket card.
I am sure both are very similar

A debayer module would not connect to computer like the card so it would not enhance any NLE performance. No sales threatened, these are two different things.

EDIT:
ideally it would be great if the brain itself could act as a debayer module but I would make an educated guess that all brain asics are set in stone by now, therefore I suggested the 'debayer module' as a stand alone that is controlled by REDmote.

 

[Gavin Greenwalt]

EDIT:
ideally it would be great if the brain itself could act as a debayer module but I would make an educated guess that all brain asics are set in stone by now, therefore I suggested the 'debayer module' as a stand alone that is controlled by REDmote.

This is pure speculation just like my speculation (seemingly confirmed by Jim Jannard moments ago) that it does a full debayer before downrezing but my assumption would be that the module interface is no where near fast enough to handle the 1-2GBs required for a full uncompressed bayer pattern. And then you need a recording medium capable of handing 3x datarate redcode and a 3x rate compressor.

My guess is the module interface is something like USB3.0 and eSATA run in parallel. Neither of which could handle uncompressed 4k bayer patttern.

 

[J Davis]

This is pure speculation just like my speculation (seemingly confirmed by Jim Jannard moments ago) that it does a full debayer before downrezing but my assumption would be that the module interface is no where near fast enough to handle the 1-2GBs required for a full uncompressed bayer pattern. And then you need a recording medium capable of handing 3x datarate redcode and a 3x rate compressor.

My guess is the module interface is something like USB3.0 and eSATA run in parallel. Neither of which could handle uncompressed 4k bayer patttern.

That is a valid point if it were connected to a camera and functioning in realtime. But it isn't, just in/out to media. No need to be in sync with anything just running at whatever speed it runs at. When viewed in that light, would it be safe to assume the module interface speed would not matter?

 

[Gavin Greenwalt]

That is a valid point if it were connected to a camera and functioning in realtime. But it isn't, just in/out to media. No need to be in sync with anything just running at whatever speed it runs at. When viewed in that light, would it be safe to assume the module interface speed would not matter?

Sure if it didn't need to output in real time and you don't mind outputting uncompressed 4k or double compressed redcode. But by this point what's the advantage to that over a laptop *not attached to the camera*?

This would make a much better module for the RED Station since it's all about offline Media IO.

 

[Nick Shaw]

My guess is the module interface is something like USB3.0 and eSATA run in parallel. Neither of which could handle uncompressed 4k bayer patttern.

I may be imagining it, but I thought it had been suggested by Red that the module interface is PCIe.