Does AJAs IO HD solve REDs poor down conversion from 4K to SD?

[laguun]

For those really interested in the subject, i published an series in the IDG magazine "digital production", i think it was 99/00/01, called: "The triangle of beauty and quality: Speed, sharpness, silkyness.

This article explained quite en detail what happens behind the scenes when scaling images, how to judge different interpolators/bicubic filters etc as well how the different systems and softwares from discreet, avid, quantel etc handle this.

I am not sure if its still available, however it covers exactly the discussed topics.

 

[Carl Dieker]

Thanks for all the replies!

Thank you, Matthew, for your post.


Perhaps Red's new format which scales more directly to 1080 will achieve acceptable down-conversion via software. Just speculation; I'm not an engineer.

Leo

Leo, are you talking here about the 1080 playback format already presented in build 16 or something more advanced that will come in 17?

Yes, I am a little bit confused by some of the posted replies when coming to dismissing taping the output feed (not the original RED CODE RAW 4K) from RED and down converting it to SD. Are people talking about the simple preview out mode (720p) or are they referring to the recently introduced 1080 playback mode?

Just to be sure that there are no misunderstandings, my question is about if REDs current playback output mode of 1080 (cropped and down converted from 4K in camera) change or add anything to that interesting discussion that was originated by Leo and others last May?

I am really looking forward to see if my Danish neighbour Tony alias "Tlorenzo" can achieve what x86box already is claiming to be possible, the kind of down converted SD one would expect coming out of a RED.


/Calle


PS: Myself I am pretty satisfied with my current workflow of pro res 1080 and with the tools of FCP, RED CINE and CRIMSON. But as pointed out before, there are people/clients with complex shots who need a solid SD workflow/delivery or else they go and rent something less spectacular than RED.

 

[Carl Dieker]

With Build 17, we're getting a 4XHD record mode that is 3840x2160. It's a bit smaller than 4K, but will give us clean down-scales to 1080p as well as un-cropped playback. Although, it is cropped from the 4K area at the time you frame and record your shot.

:sorcerer:

Maybe with build 17 aja IO/HD will make 4K down converted to SD work like a charm?

 

[Mark Crabtree]

HD means High Definition

HD means High Definition

:sorcerer:

Maybe with build 17 aja IO/HD will make 4K down converted to SD work like a charm?[/QUOTE

AJA is not going to be a solution to this problem because it does not have the necessary circuitry.

This down conversion problem is NOT Red specific. It will happen with any HD file. Standard def is garbage. Making good standard def images has long been considered more of an art than a science by the engineers at companies like Ikegami, Panasonic and Sony. We all need to wake up and smell the coffee. If you want SD that looks as good as 35mm film then you are going to have to take the same route to SD that film does. HARDWARE. HD decks work, Teranex works.

 

[Ed Watkins]

:sorcerer:

Maybe with build 17 aja IO/HD will make 4K down converted to SD work like a charm?

I just want to re-state this:
The problem does not lie with the RED footage per se; it lies with the software down conversion and interpolation methods used to create low resolution formats.

I don't see why RED keep getting blamed for this. If you want the highest quality down convert you need to use a hardware based solution. That has always been the case.

I agree that a better software solution is needed, but this is where 3rd parties with the RED SDK should be stepping up to the plate. IMHO.

 

[Harrison Diamond]

Leo, until the Quad HD format is added in build 17, perhaps you could attempt a crop (not a scale) in Redcine to 3840x2160 on a test clip, then try the scaling to SD? That should at least give us an idea of how it would work.

 

[Matthew Verkler]

The Quad HD function is primarily useful for going down to 1080 HD as cleanly as possible, as it is exactly twice the pixel dimension in each axis. That means going to 1080 will be exactly half the size in each axis, which makes for the cleanest interpolation. Digital scaling benefits greatly from round factor numbers; anything that isn't round, such as going down to a size of .55, for example, is difficult. The closer you get to a round number (without becoming a round number), the harder it is for digital downscaling. Take a look in Photoshop, for example. When you are viewing a pic at 50% size, it looks great, even though it's being scaled. Now try it at 66% or 33%. It's aliasing and crunchy, because its interpolation is not up to doing these off-factor scales. 25% is good. Even the highest quality interpolator will benefit from scaling by a factor of 2 and so forth.

So SD downconversions shouldn't benefit too much from the Quad HD format, except that if they are being downconverted from a 1080 intermediate step, the 1080 itself should be better to begin with. I think that 1080 output live from the camera should benefit GREATLY, as it will be a much easier downscale for the finite Red processing power on board the camera.

On another topic, some recent posters have reiterated my earlier point, that the Red format itself is not a barrier to good downscaling. Quite right. Leo makes the point that many people's Red workflow doesn't include nice hardware downscaling, and that there is a real need for better quality (and faster) downscaling in software. Also quite right. There is one area that makes Red footage more difficult to downscale: it starts at a much higher resolution than HD and other formats, and so is more prone to aliasing when downscaling to SD with a questionable process. But with good interpolation, having more pixels to start with will produce smoother images while remaining sharp for the destination resolution. Definitely a Red advantage.

P.S. By round factor numbers, I mean 1, 2, 4, 8 and so on. Which means scaling down to 50%, 25%, 12.5%, etc.

 

[conrad gaunt]

Thank you, Matthew, for your post.

Here's my view of why Red presents a down-convert problem that other formats do not suffer.

When we shoot 35MM film, the footage is "down-converted" either in telecine or via the HD deck's built-in hardware. Both produce very good results. In other words, no problem exists; we get alias free footage in real time and can then edit with no problems.

Shooting traditional 2/3" HD is equally seamless; the decks handle the down-conversion beautifully.

Red on the other hand, presents a serious problem to those who don't possess HD decks or dedicated, and relatively expensive, hardware.

This comes as a most unpleasant surprise to many Red producers; they shoot, edit beautiful HD images, then attempt to down-convert with an AJA card or software such as FCS or Compressor with unacceptable results.

To date, no software solution has been found to be acceptable. AJA Kona can't cut it, Scratch can't do it; Terranex works, HD decks work. Other hardware may work as well, but I've been very disappointed with a couple of them. Out of the box, test, pack it right back up and return. No, I won't name them.

Perhaps Red's new format which scales more directly to 1080 will achieve acceptable down-conversion via software. Just speculation; I'm not an engineer.

I'd love to see Red address this issue, either with a software solution that works, and works fast enough to be practical, or a hardware device.

I'm using the HD deck solution which works well.

Best regards,

Leo

I have no problem downsampling 4k straight to sd. Converted a 4k clip of a spider in a web straight to 720x405 bitmaps yesterday using my own downsampler, which I can't be bothered to provide a link for anymore because no one was interested, and it looks amazing, all web threads present and beautifully anti-aliased even at subpixel levels. You just can't beat a floating point non-realtime software downsample of 16bit Tiffs... Or.. use a crap/expensive realtime card if you prefer speed over quality, or some speed optimized software (then moan about the poor results). No offense, but downsampling 4k red footage isn't just easy, its a pure pleasure.
As for monitors not showing up problems.. well, its not hard to adjust brightness and contrast and see whatever jaggies are there, even on a bad monitor...

Leo, I did PM you about this and you didn't follow up. I've been downsampling 3k and 4k CFA footage for years in the form of stop motion, programming and dealing with discrete samples for much longer..
Please post the most problematic 4k jpeg/png you have and I'll downsample it to any resolution you wish (odd or even) and post it straight back to you.. if you really have problem?

Thank you Leo for another honest post concerning to this utterly maddening dilemma. Look I use the Red as much as anyone else here. WHY IN THE WORLD should this be so difficult an issue? Nobody wants to degrade these wonderful images that come out of our Red cameras, and yes, I understand that the images in SD wont we nearly as good as the original 4k images. What I don't see is why there isn't a simple fix to this. An AJA Kona Card? OK, but For Standard? I could understand if I were going out to Bluray even HD-DVD, but WHY should I invest thousands more for STANDARD? Ok let me make is SIMPLE, for all the experts who don't understand what I mean or why I'm complaining: HOW DO I GET RID OF THE UGLY JAGGED LINES IN SD? it's not an art, it shouldn't be rocket science, it's a simple question. Can somebody finally answer it?

I believe I can answer that. A good downsampler (for SD conversion from 4k moving images), will leave the resulting image anti-aliased (ie, not over-sharpened, but details preserved), which is also great for lossy/compressed footage, with as little noise added by accumulated rounding errors as possible.. then add a little sharpen to taste if you like, just like Red advise with the original footage.

Forget realtime, downsampling should be a final mastering step, just before encoding and outputting to a SD monitor, again, my opinion.

Another alternative is using a Sony PD150 instead of a Red One

Photoshop CS for example, has my least favourite resampling, and seems to have some kind of adaptive edge-sensing going on (there are several types of bi-cubic available in CS), but they're not meant for moving images. It can actually make images that look good for still shots.. but were trying to downsampling moving images..

 

[Joe Walker]

I just want to re-state this:
The problem does not lie with the RED footage per se; it lies with the software down conversion and interpolation methods used to create low resolution formats.

I don't see why RED keep getting blamed for this. If you want the highest quality down convert you need to use a hardware based solution. That has always been the case.

I agree that a better software solution is needed, but this is where 3rd parties with the RED SDK should be stepping up to the plate. IMHO.

Don't think anybody has been attacking RED over this, at least I hope that's not the message we've all been pushing here on this issue. Right now we're all just curious what the best methods are, and so far Leo and I both have only found two hardware methods that do the job adequately.

Leo, I did PM you about this and you didn't follow up. I've been downsampling 3k and 4k CFA footage for years in the form of stop motion, programming and dealing with discrete samples for much longer..
Please post the most problematic 4k jpeg/png you have and I'll downsample it to any resolution you wish (odd or even) and post it straight back to you.. if you really have problem?

Hey Ronx, is your program Windows only? Do you have a mac version as well?

 

[Donald G]

Spooky......

Spooky......

Another alternative is using a Sony PD150 instead of a Red One

Dude........How in the WORLD did you know that I own a sonyPD150?????hmy: Lucky Guess..... I Hope!


in any case, I would DEFINITELY be interested in trying out the downsampler that you use. Please post the link or PM we with the info.

 

[David Birdy]

We are using a "Box technologies" CPU and Scratch to output real-time 2k images at 100-120 frames.
The output of the CPU is connected to a Sony SRW machine, we make one dub of HD & use the SRWs' down-converter out to a Digi-beta deck for the SDI copy.

Let me know if you need some conversions done in this fashion, I may be able to book some overnight dub time.

Dave

 

[Alex Carr]

Wait, is the issue from the 1080 playback into a Downconverter from the camera only? I dont think you will ever get good results from that. They are similar to proxy quality... We recorded some footage on a Archos(I think) portable recorder on the Steadicam rig, the 720 on B15 went through Aja's HD-SDI to NTSC. Not fantastic, but great for Dailies... The real use was playback for the operator.

Onto the Software side:

More of what I'm saying is Bitrate, Dv-Ntsc is 25, DV50 is 50, Uncompressed is 300-400. If you are not using Uncompressed then your problem is solved. Compression creates aliasing. Stop compressing and you'll see less of the Aliasing. Once you define the best Filter for export then you have found your personal taste. This issue is directly related to operator. It has nothing to do with Red, all digital images have pixels... Push the pixels into the right place! And TEST TEST TEST!!!

I have no problems, I agree with Laguun.

 

[conrad gaunt]

Hey Ronx, is your program Windows only? Do you have a mac version as well?

Yes, but it works under Linux using Wine also, or a Mac equivilant. It uses no APIs. I was hoping to port it to my new iMac, but it isn't as easy as I thought. I'll pm a link when I've checked the website, i think I' may have made changes to the software and not uploaded them yet.

 

[Joe Walker]

Onto the Software side:

More of what I'm saying is Bitrate, Dv-Ntsc is 25, DV50 is 50, Uncompressed is 300-400. If you are not using Uncompressed then your problem is solved. Compression creates aliasing. Stop compressing and you'll see less of the Aliasing. Once you define the best Filter for export then you have found your personal taste. This issue is directly related to operator. It has nothing to do with Red, all digital images have pixels... Push the pixels into the right place! And TEST TEST TEST!!!

FWIW, we've always used Uncompressed, nothing else, and in certain high contrast high detail shots, we're still seeing aliasing once we've bumped down to NTSC. And again, we're not blaming RED for this, if anything, we're blaming NTSC. We love the images this puppy gives us, its great. We're just trying to find a more viable way to get good downconverts of this footage so all these people with NTSC TV's can admire it with us!

 

[Tony Lorentzen]

Initial tests are very promising... The IO HD performs great - this will be a real lifesaver if you need to convert to 1080p ProResHQ fast. More tests tomorrow will show scaling/aliasing issues (if present).

 

[Raphael Varandas]

Het Ronx let me know this mac soluition...pm if you could too

thanx

 

[Domenic Mastrippolito]

testing the aja IO

testing the aja IO

tlorenzo
had a chance to do any testing...any results?
Sparky

 

[blakerblaker211]

Great. Yet another thread with posters claiming that there are no problems down-converting Red footage to SD.

Let me state flatly that the AJA down-conversion is unacceptable. The usual disclaimer; not a knock on AJA, just this specific issue. Anyone who says otherwise has zero credibility with me. Sorry, nothing personal, but the only advice I'd be likely to take from them is to duck in a gun fight.

Maybe those claiming acceptable software results are viewing on monitors that hide the problems. Hopefully it's not because they don't know a jagged edge when they see one.

I don't want to flame anyone, but anyone claiming that the problems exists only for the incompetent are just plain mistaken.

Down-conversion should not be some black art; it should be straight-forward and fast enough to be practical. No one, absolutely no one, has been able to demonstrate to me a satisfactory software down-conversion. Period. I had a lot of people claim to be able to do it, but none were able to produce results even approaching those obtained by dedicated hardware.

By the way, it's not just a need for speed; no software solution works. Really smart guys have suggested doing a bit of pre blur. Fine, if you want your footage to look worse than BetaCam original; it's not acceptable for me.

If you want to down-convert Red footage without jagged edges, the only solution I've found after EXTENSIVE search is hardware. End of story.

I apologize in advance if I've offended anyone; it's not my intent. I'd love to have a fast, effective software solution to a very vital requirement, but have found none.

Best regards to all,

Leo

So what Hardware would you suggest?

 

[Dick Maney]

Leo,
Thank you for continuing to hammer away at this issue of aliasing, etc. in standard def motion pictures derived from HD. I have tried a combination of gaussian blur and sharpening, and although it helps, there has to be a better way. I agree it would be great to see a good software based solution to this problem. Meantime, we'll just limp along.
-Dick