4K is too much data and 35mm FOV is good.

[I Bloom]

Sorry this is yet another thread on this topic, I just think myself and others need to clear our heads. I want to try look at the discussion in a different light.

1080pRGB scaled from 4K does not make good sense.
720pRGB scaled from 2K does not make good sense.

We loose a lot and gain very little and we don't get any closer to a cross compatible deliverable. I'm convinced.

If we loose these formats, what are we missing: an economy format.

The discussion got confused. Because many of us believed that there was a savings by switching down to an HD scale. Unfortunately I think the only one that actually made much sense is 720p Scaled down from 4K, which I'm coming to think is not part of the spec anymore anyways, maybe it never was and I just don't know how to read the chart.

Once you've recorded the footage then as far as I can tell your choices are:

1. Find clients who cut in Final Cut Pro 6
2. Transcode in REDCine and hand over a deliverable.
3. Pull tape off the HD-SDI

So, if transcoding is ubiquitous I'm coming to think that the discussion is really about whether there is any format in the spec that dramatically cuts down on the amount of data while keeping many of the qualities that the camera offers, i.e. high dynamic range, selective focus.

In other words can we do some REDCine in the camera in situations where 4K or 2K is too much.

That's just how I'm seeing it now. I'd like to see another chart on RED.com.

A made a diagram to try and describe what I'm talking about. The arrows coming out of the boxes are meant to indicate a spectrum of possible bitrates that might be desired.

http://www.reduser.net/forum/uploaded/582_1186090971.jpg

[Dexter Gregoire]

As I understand it:

1. It doesn't matter what Software your clients edit in. Every software has a limit and they will tell you what format they want it in.

2. Transcode in REDCINE the requested format and hand over the deliverable, Backup footage on Dual-Layer DVD's - 8.5Gigs. Free up your hard drive.

The only difficulty I see maybe is transcode now. Hand it over, I'm waiting and the clients foot is tapping the ground as the computer is working. One more minute.

[Häakon]

If you want 35mm DOF and you want RAW capability, there's only one choice: 4K.

You're right, 4K is a decent amount of data. But scaling to 1080 isn't very efficient either because now you're in the RGB space and that's gobbling up three times as much bandwidth as RAW. You're nearly back to where you started, so you might as well just shoot 4K in the first place. The reason to shoot 1080p RGB, however, is expanded framerate.

The basic fact here is that in order to get 35mm DOF, you need to be shooting with a 35mm sensor. That's what this part of the camera is about. It's hard to make an "economy" format that has the DOF you want because the simple truth is that what you're really after is 4K worth of data. 4K is big, that's just something that has to be dealt with. Fortunately, REDCODE RAW has taken that uncompressed 4K (which is close to 1GB/s) and made it manageable for a heck of a lot more users in a nice 27MB/s package. Getting it smaller than that is like saying you want your 12MP dSLR to shoot full-frame 50kb stills. Just not gonna happen.

[I Bloom]

If you want 35mm DOF and you want RAW capability, there's only one choice: 4K.

You're right, 4K is a decent amount of data. But scaling to 1080 isn't very efficient either because now you're in the RGB space and that's gobbling up three times as much bandwidth as RAW. You're nearly back to where you started, so you might as well just shoot 4K in the first place. The reason to shoot 1080p RGB, however, is expanded framerate.

The basic fact here is that in order to get 35mm DOF, you need to be shooting with a 35mm sensor. That's what this part of the camera is about. It's hard to make an "economy" format that has the DOF you want because the simple truth is that what you're really after is 4K worth of data. 4K is big, that's just something that has to be dealt with. Fortunately, REDCODE RAW has taken that uncompressed 4K (which is close to 1GB/s) and made it manageable for a heck of a lot more users in a nice 27MB/s package. Getting it smaller than that is like saying you want your 12MP dSLR to shoot full-frame 50kb stills. Just not gonna happen.

I think I make it pretty clear in my post that 1080pRGB is not an economy format, and so has no place in the spectrum that I'm seeing. I don't think I'm after 4K worth of data, I think I'm after the depth of field characteristics inherent in large sensor and other characteristic while otherwise willing to tolerate more compression artifacts, for example, because I know the image will be transcoded to 720p DVCPRO HD anyways.

I can't find the post but someone on the RED team, I think it might have been Rob, acknowledged that working with 4K files is difficult because of the size. We might find that we have some clients who can deal with a RED codec and will cut in FCP6 but just don't want that glut of data that they will never need.

I'm suggesting that the gap opened up by removing 1080pRGB be filled by something that actually accomplishes what 1080pRGB was meant to accomplish, which is a quality compromise in camera that reduces data and processing load while still using the whole chip.

IBloom

[Häakon]

I think I make it pretty clear in my post that 1080pRGB is not an economy format, and so has no place in the spectrum that I'm seeing.
I agree completely, I said, "scaling to 1080 isn't very efficient either because now you're in the RGB space and that's gobbling up three times as much bandwidth as RAW. You're nearly back to where you started, so you might as well just shoot 4K in the first place."


I don't think I'm after 4K worth of data, I think I'm after the depth of field characteristics inherent in large sensor and other characteristic...
My point was that the only way to get the full-sensor data characteristics you seek is to either 1) shoot with the full sensor (which means shooting in 4K), or 2) scale the data (which is what they've provided us in the forms of 720p and 1080p).

You can't have RAW scaled data, so the only way to present the 720p and 1080p data is as RGB - and along with that comes higher file sizes (which, again, keeps this from being an "economy" format like we've both agreed upon).

Therefore, there is no way to have 4K sensor data at "small" filesizes without severe compression, and most users who have a need for/are willing to pay for a 35mm sensor in the first place are not going to find a quality hit of that magnitude acceptable for their master footage.


I can't find the post but someone on the RED team, I think it might have been Rob, acknowledged that working with 4K files is difficult because of the size.
They have remedied that largely by providing the REDQUICK component, which will allow for near-instant and seamless conversion of your 4K master into very manageable filesizes (2K, 1K, or .5K) without having to touch your original "digital negatives." And if you need to compress your footage even more, you can certainly transcode it to whatever format you wish in REDCINE. More information about REDQUICK can be found in this topic:

http://www.reduser.net/forum/showthread.php?t=3501


I'm suggesting that the gap opened up by removing 1080pRGB be filled by something that actually accomplishes what 1080pRGB was meant to accomplish, which is a quality compromise in camera that reduces data and processing load while still using the whole chip.
I'm suggesting that the answer for maintaining full-sensor characteristics at manageable filesizes has already been provided to us in the form of REDCODE RAW, REDQUICK, and REDCINE. The only reason I see to maintain keeping RGB (and for shooting it) is for higher fps while maintaining full-sensor readout (which is both justifiable and important).

[Gavin Greenwalt]

Shoot 4k.
Convert to 720p DVCPro HD straight off the extra REDDrive or CFCard while capturing.
Copy DVCPro HD file to Client Drive.
Deliver.
Profit.

If your client can't take REDCode in any form... then the camera shooting 720p only slightly extends your record time and you'll still have to transcode to DVCPro or DNxHD or whatever format they need it in. End of story.

If 4k is too much data for you to store: transcode to 720p 10bit log RGB REDCode with REDCine for yourself and delete the original RAW file.

[Andrew Young]

You're right, 4K is a decent amount of data. But scaling to 1080 isn't very efficient either because now you're in the RGB space and that's gobbling up three times as much bandwidth as RAW. You're nearly back to where you started, so you might as well just shoot 4K in the first place. The reason to shoot 1080p RGB, however, is expanded framerate.

I think Haakon is right here. iBlooms desire for an economy format is totally logical, but we're probably asking for a format (scaled RAW), that is not technically feasible. Now that the inefficiencies of 1080p RGB are more clear, and the need to transcode remains, the value of the format lies solely in frame rate, which is still a valuable consideration - unless significant frame rate gains can be made in the 2K/4K formats.

If your client wants DVCPro720p and disk economy is important, it probably makes sense to shoot windowed 2K and transcode from that. And if your client insists on 35mm DOF, then why is he working in 720p? He's either got to deal with lower disk economy or accept a 16mm DOF, which is still better than what he would get if you were shooting with a 2/3" camera.

What we REALLY need to know at this point, as many have stated, is HOW LONG DO TRANSCODES TAKE?!

[I Bloom]

I think you are overestimating what "severe" compression would be.

Here is an example of the type of format I'm looking for in the camera:

1. Capture 4K Data from the sensor and demosaic to 4K RGB:
2. Scale to 2K RGB
3. Remove 2/3 of the samples, creating a "soft" 2K Bayer Pattern.
4. Compress using REDCODE RAW.

There you have it, a 7MB/second format, in camera, using the whole chip with the same image quality as 2K Windowed. Voila!

IBloom

[Andrew Young]

Here is an example of the type of format I'm looking for in the camera:

1. Capture 4K Data from the sensor:
2. Scale to 2K or 1080p. RGB
3. Remove samples, creating a "soft" Bayer Pattern (the programmer would really just not store the samples)
4. Compress using REDCODE RAW.

There you have it, a 7MB/second format, in camera, using the whole chip with the same image quality as 2K Windowed. Voila!

IBloom

Given the advantages of RAW this would be fantastic, if it were possible. The question is can you scale the RAW without decoding to RGB. I'm guessing this would have been offered if it were possible.

[Dexter Gregoire]

There you have it, a 7MB/second format, in camera, using the whole chip with the same image quality as 2K Windowed. Voila!

IBloom

That would be great if possible. We have to remember that we had been warned time and time again from many different companies - High Definition takes up alot of space. Even Avid & FCP came up with their DNxHD versions to handle the large data issue. RED has offered 4K at around 27 MB/second and you've presented something around 7MB/second.

The program is probably already written, I would prefer seeing the camera go to production and get mine, rather than have programmers modify their program and test your theory. That takes alot of time.

I think at the end of the day Hard drives are not very expensive. I'm prepared to buy a few more. That's the cost of 4K or 2K or HD.

[Gavin Greenwalt]

How are people affording a $30,000 kit and hiring a good focus puller but not able to afford a handful of hard drives? We're talking $100 a day for 5 hours of footage. 4k is already economy. I can't believe I'm saying it, but it's true.

4k RAW - 27MBs 12 bit
2k RAW - ~ 7 MBs? 12 bit
1080p RGB - ~ 21 MBs 12bit
720p RGB - ~ 9.33 MBs 12bit

So that means if you shoot 5 hours of footage you can pick up the $69 HDD for the day instead of the $100 HDD by shooting 720p RGB.

Congradulations your 'economy' option has saved you.... $31 a day and in the process reduced your negative in quality by a factor 16. You still have to transcode to DVCProHD or equivalent if you want to hand them a non-REDCode deliverable. You still have to copy it off the REDDrive of CF Card and onto a HDD for the client.

720p is economy like the car that cost $100 but burns a quart of oil every week.

[I Bloom]

Given the advantages of RAW this would be fantastic, if it were possible. The question is can you scale the RAW without decoding to RGB. I'm guessing this would have been offered if it were possible.

It´s possible, most of the neccessary steps were already being implemented for 1080pRGB scaled from 4K. One step I left out is that you first interpolate the 4K Bayer, which as Graeme said was already part of the plan.

The only thing I´ve added is dumping samples and going back to a Bayer Pattern. You are essentially getting the exact same image as 2K windowed only you started from a larger field of view.

You can at that point , I believe, use the same compression, you would use for 2K windowed. You only have the additional overhead of scaling.

IBloom

[Jeff Kilgroe]

I think the problem with scaling RAW and retaining usable RAW data rather than converting to RGB is that the entire bayer area to be scaled (like the 4K sensor area) would have to go through a demosaic, then get scaled, then re-quantized into a 12bpp bayer pattern.

I don't think it's impossible... I think it's more of a question of what the internal processors of the RED One can handle at usable frame rates.

[I Bloom]

I think the problem with scaling RAW and retaining usable RAW data rather than converting to RGB is that the entire bayer area to be scaled (like the 4K sensor area) would have to go through a demosaic, then get scaled, then re-quantized into a 12bpp bayer pattern.

I don't think it's impossible... I think it's more of a question of what the internal processors of the RED One can handle at usable frame rates.

Jeff,

A while back Graeme told me that this is already happening, read below:


We do full debayer and downscale in camera for the RGB recording pathway for best quality. For RAW recording, it's done in post.

Graeme


Do you debayer to 4K and then downscale?


In camera, yes.

Graeme

So consider again what I'm proposing:

1. Capture 4K Data from the sensor and demosaic to 4K RGB.
Already happening!
2. Scale to 2K RGB
Already happening!
3. Remove 2/3 of the samples, creating a "soft" 2K Bayer Pattern.
An extremely simple program!
4. Compress using REDCODE RAW.
Already happening!

Basically what I'm proposing is the Bayer pattern equivalent of 2:1:1 color sampling from an image that could be 4:4:4. To me it seems simple to do and valuable. It fits elegantly into RED's existing pipeline.

IBloom

[Häakon]

So consider again what I'm proposing:

1. Capture 4K Data from the sensor and demosaic to 4K RGB.

Once you've demosaic'ed to RGB, you cannot go back to RAW. You can compress the footage again, using the REDCODE codec, but you've already made your RGB conversion so that step would be pointless. It would be like taking your DV footage in Final Cut and exporting it to DVCPRO HD. Sure, it's possible, but it won't magically make your footage HD quality again just because you're using an HD codec.

[Gavin Greenwalt]

Well 'can't' is a strong word. You 'can' do it. They won't give you any tools to do it, but it is possible. Just not advisable or prudent.

[I Bloom]

Once you've demosaic'ed to RGB, you cannot go back to RAW.
In fact you can go back to RAW in the sense that RAW means Bayer Pattern. You simply forget 2/3 of the samples (data you don't need). The resulting image is equivalent to the data structure of 2K Windowed RAW.

When algorithm designers test new demosaicing algorithms they do exactly what I'm talking about, they take an RGB image and remove 2/3 of the samples leaving a "soft" Bayer pattern. Then they try to recreate the RGB image with their algorithm. They then compare the two to see how accurate their algorithm is.



You can compress the footage again, using the REDCODE codec, but you've already made your RGB conversion so that step would be pointless.

REDCODE i.e. wavelet compression is the last step of the process. It's not happening again, it's happening once.

Remember the point of this is to get an image that is comparable to say an F900 with a pro35 adapter while storing 1/4 of the data of 4K RAW.



It would be like taking your DV footage in Final Cut and exporting it to DVCPRO HD. Sure, it's possible, but it won't magically make your footage HD quality again just because you're using an HD codec.

I'm pretty sure its nothing like that. Unless your goal is to go back to 4K. People using this format are using it because they want less data for finishing to HD and SD for broadcast/ENG/EFP and straight to DVD films.

What you end up with is exactly the same as 2K Windowed but with a 35mm FOV.

The Bayer pattern is a kind of natural compression. It relies on the fact that most of the missing sample data can be recreated by smart algorithms that utilize some of the common properties of natural images (largely that hue changes slowly while luminance changes rapidly).

I'm suggesting that RED could reuse this natural compression with scaled images, the same way that they use it straight off the chip, in order to create a 7MB/sec format with a 35mm field of view.

IBloom