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 Scaled Is Vital

From a processing standpoint at normal 4K Redcode, the pixels are coming in anyway in an ordered fashion and compressed at, currently, up to 25fps.

Now, with scaled Raw, the same 4K pixels come in in their ordered fashion, and a semi simple matrix manipulation is performed to turn an 8x8 pixel into a 4x4. Now you have a 2K ordered pixel array that can be put through the same 2K compression that currently exists that is now running up to 80fps.

I would think that simple offset averaging would be very simple and fast. Come on--"visually lossless compression" flies and this is nothing compared to that. Maybe not the full 60fps the sensor is capable of but certainly more than 25fps.

As Brook said the moiré charactaristics of shifting the pixels may be a problem, but then again--maybe not.

If the sensor is capable of 120fps 2K Raw, but the overhead of compression brings it to down to a current max of 80fps, then a max 60fps scaled 2K Raw should get at least 40fps if if was a linear relationship of Raw FPS to compressed FPS, but I think the slower the fps the closer to real time it would be so I think it could get much higher.
 
Thom said:
From a processing standpoint at normal 4K Redcode, the pixels are coming in anyway in an ordered fashion and compressed at, currently, up to 25fps.

Now, with scaled Raw, the same 4K pixels come in in their ordered fashion, and a semi simple matrix manipulation is performed to turn an 8x8 pixel into a 4x4. Now you have a 2K ordered pixel array that can be put through the same 2K compression that currently exists that is now running up to 80fps.

I would think that simple offset averaging would be very simple and fast. Come on--"visually lossless compression" flies and this is nothing compared to that. Maybe not the full 60fps the sensor is capable of but certainly more than 25fps.

As Brook said the moiré charactaristics of shifting the pixels may be a problem, but then again--maybe not.

If the sensor is capable of 120fps 2K Raw, but the overhead of compression brings it to down to a current max of 80fps, then a max 60fps scaled 2K Raw should get at least 40fps if if was a linear relationship of Raw FPS to compressed FPS, but I think the slower the fps the closer to real time it would be so I think it could get much higher.
Click to expand...

I definitely believe the "visually lossless compression" is far most complicated than downscaling pixels. But in this case the processor has to do BOTH - downscale AND compress. At twice the speed.

Yes, it may be quite fine and it might not. We don't know.

Thom said:
...Also, remember, there are other factors limiting the 80fps of 2K such as write speed/bandwidth.
Click to expand...

The write bandwidth at 2K 60fps is just half the one of 4K 30fps
 
dalife said:
Not only it isn't simple, but may be quite a computing load at 60fps.
Click to expand...

I've wondered that too... If the scaling of 4K RAW down to a 2K "RAW" approximation and then encoding that to REDCODE is about as computationally expensive as the processing of 4K to REDCODE, then what would be the point? I hope that's not the case.
 
2K scaled 'proxy'

2K scaled 'proxy'

Without knowing much about the real technical issues involved, I think Thom suggestion is not practical unless the RED one has HUGE processing power to be able to analyze and integrate new pixels values from an oversampled 4K image @ 60 fps... just to be able to satisfy the CF card / RED drive bottleneck.

a more 'real life' 2K scaled image (with 35mm depth of field @ 2K "windowed" datarate) would be to use only 1/4 of the sensor data (I suppose they do a similar operation when creating an instant 2K QT proxy)...
That would still requires some processing power to be able to receive 4K@60fps and deliver 2K@60fps (and maybe it is not possible to do while keeping the RAW format etc) but it seems much more a realistic "in camera" 2K scaled model (you do loose the nice 'oversampling' from Thom model) :
 
stricko108 said:
Hans,
Quick question for you, when you say shoot 50fps for 25i
would you run the 50 fps through some process to get
25i, or would you just bring your 50 fps footage into
a 25i timeline and make sure the 50 fps footage is being
scene by the editing programme as 50, so its not playing at
half speed.

Please let me know if my question is not clear, and I appreciate
any response.

Thanks
Stricko
Click to expand...

Sorry for replying so late but I had to bring my kids to bed...

Yes, you shoot 50fps, put the 50p footage on a 25fps/50i timeline and speed the whole thing up (200%). You've got now PAL20fps interlaced footage.

Hans
 
mikael said:
Without knowing much about the real technical issues involved, I think Thom suggestion is not practical unless the RED one has HUGE processing power to be able to analyze and integrate new pixels values from an oversampled 4K image @ 60 fps... just to be able to satisfy the CF card / RED drive bottleneck.

a more 'real life' 2K scaled image (with 35mm depth of field @ 2K "windowed" datarate) would be to use only 1/4 of the sensor data (I suppose they do a similar operation when creating an instant 2K QT proxy)...
That would still requires some processing power to be able to receive 4K@60fps and deliver 2K@60fps (and maybe it is not possible to do while keeping the RAW format etc) but it seems much more a realistic "in camera" 2K scaled model (you do loose the nice 'oversampling' from Thom model) :
Click to expand...

I think the Redraw compression that is being done in real time is much more complicated than the scaling Thom suggested, so it wouldnt need a lot of computing power at all. Maybe Thom's scaling method could work even faster if he didn't use 5 pixels for each green pixel but only 4.

To just throw away 3 of the 4 sections would make it look nasty. Try scaling a detailed shot (preferrably with a lot of pattern) in Photoshop with 'nearest neighbour' interpolation. You'll get an idea of what this method result will look like.
 
I'm a bit confused. When we were discussing high frame rate recording some months ago Jim was very confident that 2K scaled would happen and the argument was about what fps it would max out at: 75 or 100 I seem to remember..
I have been assuming that we would be using 2K scaled to capture 40 minutes of footage to a 16GB flash card, for broadcast or SD work. So 2K scaled is essential for flash to be useful.
Jim was very confident about it, so I'm sure it will happen eventually.
 
fergus said:
I'm a bit confused. When we were discussing high frame rate recording some months ago Jim was very confident that 2K scaled would happen and the argument was about what fps it would max out at: 75 or 100 I seem to remember..
Click to expand...
That'd be windowed - the sensor maxes out at 60fps for the 35mm part of the sensor, so that's a hard limit on 2k scaled.

I agree, though, that it's very important functionality for a lot of people - I would certainly use it regularly and would love it to be available...
 
Floris Liesker said:
I think the Redraw compression that is being done in real time is much more complicated than the scaling Thom suggested, so it wouldnt need a lot of computing power at all. Maybe Thom's scaling method could work even faster if he didn't use 5 pixels for each green pixel but only 4.
Click to expand...

Thinking what they were planning to do, 1080p, RGB at higher speeds, not only would they have to scale and compress they would have to debayer as well, I really think that simple single number averaging is really no cycles as compared to debayering.

I have no doubt in my mind the processor can handle simple pattern averaging with little overhead, but it really goes back to the visual artifacts and patterns this may introduce.

If Graeme hasn't chimed in by now, it's because he's got something in the works--that likely involves way more math, but works 10x better and is faster than we can imagine.
 
Hans von Sonntag said:
This is a R&D request that I humbly bring up because I came over it in a real world situation

As Matt Uhry mentioned somewhere there is some serious need for 2K scaled. Yes there is and we badly need it.
Click to expand...

I totally agree for the same exact reasons. Actually I requested this feature from way back (in RED Year, that is) in less than perfect English. I just want to add my voice here again, hoping the RED team would put more weight on our wishes.

Back then, I gave one more reason for my request. While 4K 4:4:4 is all good, but coming from 35mm film and Sony CineAlta background who has done some filmout from 24psf 4:2:2 HDCAM tape, I would say 2K is "plenty" good for 90% of what I (we?) do. All I need from RED ONE is the depth of field look and significant overcranking capabilities. High end digital cinema projector may do justice to 4K footage with richest color detail, but print stock projected from traditional film projectors? I doubt it.

Honestly, how many of us actually need to do filmout. If RED stops taking orders today and every camera they ship will make one film for theatrical release per year. There will be 3000 more movies each year, mostly from independent producers who want to make a professional looking film at lower cost. Can the market really consume that much supply (without big names attached and unbelievable production scale)?

This might not be a popular thought around here. But that's OK. I just want to speak my mind. On the other hand, RED ONE will certainly make prime time TV shows in other regions of the world look a lot better, knowing what some Asian TV drama series look like today. But that doesn't require all the data in a 4K 4:4:4 image. Besides, lower data rate would also make storage and post work a lot easier and faster, possibly eliminating offline/conform steps all together.

While on this topic, I remember reading an earlier debate in this forum about dropping certain "video" recording formats from the camera because the RED team believed (or was contemplating) that they were "not needed". I want to take this opportunity to respond by saying: please keep all the options available to us—the end users. We need choices for different occasions. Choices won't hurt anyone.

To give you an example, recently a poster in the Lens Tests forum was wondering if he should get the 12 pound Cooke 18-100mm lens, a heavy, bulky 5.5:1 cine zoom with slow aperture of T3, for his interview and corporate communication work, huh?! Quite obviously, he is not too familiar with the "film world" that RED ONE is leading us into, but he is a user after all and he deserves to have his choices.

Please, please keep all the formats in the camera and make the 2K scaled, hopefully capable of higher than 100fps if technically possible, happen soon.

Thank you in advance.
 
Thom said:
If Graeme hasn't chimed in by now, it's because he's got something in the works--that likely involves way more math, but works 10x better and is faster than we can imagine.
Click to expand...

Graeme, report in!
 
All In One,
Spot on! The scalability originally promised (subject to change) in this camera was one of its best assets. 4K, 2K, 1080, 720, 480(why?) all SCALED.
Hopefully we'll be able to boot into different modes eventually...
Cheers,
Harry
 
Thom's Bayer scaling is not quite right. His method results in unevenly spaced green pixels:

G B G B G B G B
R G R G R G R G
G B G B G B G B
R G R G R G R G
G B G B G B G B
R G R G R G R G
G B G B G B G B
R G R G R G R G

becomes:
......................
...G B......G B....
...R G......R G....
......................
......................
...G B......G B....
...R G......R G....
......................


The correct scaling averages 4 nearest green pixels to make each resulting green pixel. Here the X is the averaged result:

...G
G X G
...G....G
......G X G
.........G

So that:

...G.........G
G B G B G B G B
R G R G R G R G
...B G B G B G B G
R G R G R G R G
G B G B G B G B
R G R G R G R G
...B G B G B G B G
R....R G R....R G


results in:

..G..........G........
......B..........B.....
...R....G....R.....G.
........................
..G..........G........
......B..........B.....
...R....G....R.....G.
........................
 
harryclark said:
All In One,
Spot on! The scalability originally promised (subject to change) in this camera was one of its best assets. 4K, 2K, 1080, 720, 480(why?) all SCALED.
Hopefully we'll be able to boot into different modes eventually...
Cheers,
Harry
Click to expand...

agreed.

high frame rates with full sensor are completely essential - 1080p would be fine as far as i'm concerned.

a 35mm camera that can shoot high frame rates is much more important to me than being able to record 4k.

can we please hear from red about progress on this?

cheers,

david
 
DSPographer said:
Thom's Bayer scaling is not quite right...
Click to expand...

I certainly could be wrong, (happens almost every day! :) ) but at this point--I don't see it by your example.

The resultant image needs to be a proper Bayer pattern. As you demonstrated in showing my sample--it still looks to me like it is. As I see it, what we are trying to do is reduce a 4x4 section to a 2x2 one.

Each 4x4 square has 8 greens, 4 blues and 4 Reds, my resultant 4 pattern maintains the same ratio.

To me, as long as I represent the area with a bayer pattern with the same ratio and maintain the same bayer pattern--it should properly maintain the color representation.

When I look at your example, your resultant image doesn't seem to be a proper bayer pattern --or am I missing something?

I attached another image to explain what I'm thinking...

bayer2.png


Remember: the collapsing of the sections into the middle is just to show how the pattern is just representing a view of the overall image. The true reduction goes directly from step 1 to step 3.

I certainly enjoy this type of discussion and look forward to learning if my theory somehow corrupts the pattern of throws off the relationships.

Best Regards,

Thom
 
You must log in or register to reply here.
Back
Top