Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

2K scaling...

Status
Not open for further replies.
Scaling can have any degree of sharpness or softness depending on the downsampling filter choice, of which there is an infinite variety and no "right" choice, but always a balance between sharpness, aliasing and ringing.

Graeme
 
hi jeff, thank you for your reply... sounds good with the increased cpu efficiency . However in terms of conversion time, I think 2k scaled would be THE biggest thing coming down the pipeline, especially if one takes the improved cpu usage into consideration. These steps taken together would cut conversion times seriously. And this is really a major issue.
Imagine having seven hours of footage from an event and the editor wants dnxhd... what are you going to tell him... "fine... see you in two weeks"?!

Since all the RGB formats went down the drain, and now also the data efficeint full chip alternative. A shooter now have no posibility of delivering a couple of hours worth of red-recordings without a serious penalty... And I am talking "days of pizza, coke and videogames"?!?

This prescaling is really crucial.... am I missing something or why are people not stepping on eachothers toes here?
 
im.thatoneguy said:
portable_mainpage.gif
Click to expand...

The website says "... It can also record Red Digital Cinema’s RED ONE™ camera in 4K data-mode, when this becomes available"

Is this something new that we haven't heard about yet, perhaps using the SATA port, or is it the Raw-port option?
 
luis bustamante said:
As far as I know, it will take data from the Raw Port and convert it to cineform RAW format. It will cost US$40K aprox.
Click to expand...

Amazing price tag!

Where did you get that info from, that it will convert to cineform RAW?
I've not found any informartion like this from their website.

Anyhow, this price is really excessive.

Cheers,
Axel
 
Luis,

I think they just meant that it can export in those formats, just as RED CINE or RED ALERT would do as well. But who knows...

Axel

Cite:

[...]
It also applies LUTs (without baking), keeps all metadata (makes metadat input easy) and wraps everything up for fast delivery to Post as DPX or (with the Transfer Station) any Post deliverable you can think of (Avid MXF, QT, Cineform, AVI, JPEG etc).
[..]
 
Graeme,

In scaling to 2K raw in a balance between sharpness and artifacts is it possible to obtain an image which will appear as sharp as the windowed 2K and contain no more noticeable artifacts?

If you cannot do this then I certainly would call 2K scaled worthless and impossible.

If you can do this then it is essential for many of us who work in the corporate, doco, and reality worlds. It will also work for many dramatic projects which use a lot of high speed where the windowed 2K may better match with scaled 2K. It will also work better in many venues like the school where I teach who are not about to upgrade to the latest macs for every student desk so we can edit 4K

It's true that someday computers will be fast enough to deliver long form footage quickly enough in 4K. It's also true that a lot of us may be retired or dead of old age and the RED may be in version 7 at 12K by then.
 
And I wish I could give a meaningful answer to your question, but I cannot.

Graeme
 
tj williams said:
[..]
In scaling to 2K raw in a balance between sharpness and artifacts is it possible to obtain an image which will appear as sharp as the windowed 2K and contain no more noticeable artifacts?
[..]
Click to expand...

Not being Graeme myself, but having made tests in this field I'd say you clearly GAIN image quality by having the 2K scaled over the 2K windowed mode.

This is of mainly three reasons:

1. Averaging 4 into 1 pixels results in reduced noise levels (you can do temporal or regional sampling/averaging, and this is regional).

2. Averaging 4 into 1 pixels results in a sharper experienced image.

3. The lense issues (lines limit) is reduced to 1/4th in 2K scaled vs. 2K windowed. So if lense issues reduce resolution, the effect will be minimized to 1/4th in 2K scaled, but will be fully appearing in 2K windowed. Who knows what lense you need to use in a low budget scenario. And 2K is still a very good result with a good DOF.

The big point here is how the 2K scaled is done?

Are we talking about scaling each of the four RGBG Bayer quadrants individually, thus resulting in a new Bayer image?
This may introduce some new pixel shifting issues that might need to be addressed. We can test that out very simple by simulating this in post production.

Or are we talking about conversion from 4K RAW to 2K RGB?
This is VERY simple! We did a lot of tests in this area and after all, the simplest and usually most efficient way was to average the two green channels (rGbG) into a new one 1:1 (50%:50%) and then use the resulting 3 "Quadrants" RGB (each 2K in size, natively) as RGB image. This is extraordinary fast CPU wise and gives a very good sharpness. The pixel shift between the quadrants doesn't seem to be an big issue in our experience. But thats often a matter of taste. Technically you can even drop a green channel completely (which will result in a sharper felt image, but that is a bit complex to explain here).

I assume that the 2K quick mode in Quicktime proxy does about exactly this, so its in our hands already.
Graeme, can you elaborate on this?
This would be highly appreciated!

So, assuming I am right with the last assumption, we have a 2K scaled mode already in our hands in terms of quickly downscaling 4*2K RGBG Quadrants to 2K RGB.

But then we currently lack the reduced storage of a 2K windowed and I assume that 2K scaled looks far better than 2K windowed after all, need to test that, so intercutting the two formats may be visible. We also lack the ability to record full sensor 60 Hz. Jim forced us to ask for 4K RED CODE 60 Hz, so I raise that question here clearly once more.

Further I still believe that downscaling four 2K quadrants to four 1K quadrants and THEN doing the RED CODE compression should yield a faster processing time than doing the RED CODE on 4K directly. The math operations done in RED CODE are surely by far more complex than a bitwise add and divide by four, ie. a bit shift - extremely simple operation.

With all respect, but this seems a very valid option to me to implement the full sensor 60 Hz acquisition with what is there right now (implying that the hardware onboard can do full sensor 4K right now, out of the box).

The new DSP is said to be more powerfull. Adding this quick scale down from 4K to 2K Bayer as a third option between 4K native and 2K windowed native recording is something so many poeple here see a reason for, demanding it strongly.

I think its definetly harder to optimize the RED CODE to work in 4K fast enough if the DSP is the limiting factor (but who knows, maybe the new DSP is already fast enough and RED will surprise us all with one of the next updates).

Please don't take this offensive, just as constructive feedback.

Axel
 
Or I would venture to say that for some of us, a paid upgrade of the FPGA or DSP to enable more options, even if it means loading and rebooting (scaled 2k, direct 1080 recording to other codecs, etc. for EFP work) might be an option.
Or I'll say it again; a box, NOT tethered to the camera like the Codex box but something made by Red that you insert a CF card into, or plug a Red Drive into, with MASSIVE hardware transcoding ability.
Let's say I have a commercial client one day that wants to shoot Genesis because his post facility has a very expensive HDCAM deck they need to amortize. The next day I'm doing a promo and the company only has an in-house editorial staff with a DVCPro 50 deck. With an on-set conversion box that can crunch 4K Redcode at 2X or 4X rates into any common production codec, I can service them both, with NO laborious tutorial about the benefits of RedCode. These are real-world production needs, and I'm not sure anyone at Red has the perspective of a journeyman cinematographer whose personal business model includes renting gear to a wide variety of clients (no offense) YES, I know and you know that we can do it with a PC, but that does not convince some of my clients. One of the most popular rigs I have is my HDX-900, Pro-35, and Firestore. The client gets digital files that they can drop in and use RIGHT AWAY, as well as a free tape backup, and 35mm DOF and 1080. But I've spend a LOT of time on the phone selling this combo. Once the use it, they love it. I think for some of them, going back to a film-style finish in a broadcast environment will be perceived as a step backwards.
I know that Red is a business and they have this very modest price point to meet, but some of us would have paid more for the camera to get some extra flexibility in the feature set. Or will still pay me for a hardware "bridging" solution such as this.
But for now, let's get some cameras out the door... ;)
Cheers,
Harry
 
I've even had people decline to use my SDX-900 in favor of a Digibeta, giving up 24p in the process, because they own a Digibeta deck. Yes, it's silly. But on the more low-rent jobs, silliness prevails if the producer can make a nickel. On the commercials I have an easier time of it...
Cheers,
H.
 
Axel, you can average 4 pixels into one. You've now just used a small kernel box filter, and yes you'll get a sharp image, but probably quite aliased as well.

Similarly with the rapid techniques you describe, similar to what we use in the Quicktime codec. They're fast as you say, but as you're putting 4k content into a 2k image without using a filter, you get strong aliases, and the lack of co-sitedness of the RGB can cause issues.

Normally, broadcast quality scalers use a large polyphase filter, with lots of taps, to get the quality needed. That allows them to tweak the balance between aliasing, sharpness and ringing to a very fine degree.

Harry, your comments have been heard.

Graeme
 
Graeme Nattress said:
...
Normally, broadcast quality scalers use a large polyphase filter, with lots of taps, to get the quality needed. That allows them to tweak the balance between aliasing, sharpness and ringing to a very fine degree.

Graeme
Click to expand...

I see here the possibility for a new Red product.

Those who worry about added rendering time when downscaling might go for a processor box where you insert the CF and output to another CF or to a computer. This could be based on the current in-camera hardware and firmware.

Those who worry about storage on CF could have this unit connected to their camera and record the downscaled image. It would be bulky, but would also be able to run off its own battery and might have several CF slots.
 
"Mind the gap"

"Mind the gap"

harryclark said:
Or I would venture to say that for some of us, a paid upgrade of the FPGA or DSP to enable more options, even if it means loading and rebooting (scaled 2k, direct 1080 recording to other codecs, etc. for EFP work) might be an option.
Harry
Click to expand...

There is a gap in the HD and SD TV market for a +120fps, 720p line or greater multi format camcorder.
Eng or pro camera manufacturers are not addressing higher frame rates at the moment because they are sticking to CCDs.


regards


Mike Brennan
 
Mickjoebill said:
There is a gap in the HD and SD TV market for a +120fps, 720p line or greater multi format camcorder.
Click to expand...

Weisscam makes a pretty nice setup that shoots from 1fps to 950fps in 1280x720p. Of course, I would prefer that RED could do this so I don't have to rent one of those other super-expensive cameras.

There's also the Phantom HD.
 
Graeme Nattress said:
Axel, you can average 4 pixels into one. You've now just used a small kernel box filter, and yes you'll get a sharp image, but probably quite aliased as well.

Similarly with the rapid techniques you describe, similar to what we use in the Quicktime codec. They're fast as you say, but as you're putting 4k content into a 2k image without using a filter, you get strong aliases, and the lack of co-sitedness of the RGB can cause issues.

Normally, broadcast quality scalers use a large polyphase filter, with lots of taps, to get the quality needed. That allows them to tweak the balance between aliasing, sharpness and ringing to a very fine degree.

Graeme
Click to expand...

Graeme,

you are talking about the downscaled 4K RAW to 2K RGB issues when talking about the lack of co-sitedness?

I know what you mean, but anything you do to avoid the aliasing causes color crosstalk in one way or the othere - which might sometimes or even often be preferred, to smooth out the image.

My intend was just to point out that poeple have a very good true 2K RGB camera, when using a 4K Bayer sensor - better than a 3 CCD typically will be (for many reasons). And the process to extract this 2K can be as easy and realtime fast as this moving of the 3 or 4 color components into each other. Clearly its not absolutely correct as the original quadrants are shifted 1 pixel and have 1 pixel wide gaps - half pixels in 2K speech.

Another - potentially better - approach is to interpolate every missing 3 (or 2 for the green layers) pixels per 4 pixel RGBG and then downscale to 2K, or do a full 4K adaptive deBayer and downscale from there. So far about 4K RAW to 2K RGB.

All this is in your hands, Graeme, as you are the one who can check if the DSP can do this or that. Maybe you have done it already and we wait for the faster media to come along to cover this, see below...

The story differs when we talk about 4K RAW to 2K RAW. There the shift is much less (due to the interleaving of the pixel areas you are watching). I guess this field should be more explored and addressed. I expect that a specialized deBayer could gain some quality here, so tagging the files then as "2K RAW scaled" somehow is an essential bit of information for post.

I just remain with the question which way we can get maximum full sensor read-out at 60Hz done, and stored in an affordable and capable storage media. As pointed out many times, not everyone aims 2K finals, nor 4K finals, but the look...

But maybe the surprise will be that when RED DRIVE comes around the corner, it can already deliver 60 Hz recording full 4K, because it should be able to handle about 3-5 times better transfer rates than the CF cards. Given a much faster DSP this is well possible.

Then, only then - no more questions...

Cheers,
Axel
 
4K RAW to 2K RGB obviously makes sense except that it doesn't save much on data requirements -- in theory, 4K RAW is four-times the size of 2K RAW, and 2K RGB is three-times the size of 2K RAW... so the difference in file size between 4K RAW and 2K RGB isn't that much.
 
Status
Not open for further replies.
Back
Top