Downscaling HD -> SD Interlaced.

farss wrote on 5/31/2008, 4:42 PM
Some of us are blessed with cameras that in 24,25 or 30p deliver way more resolution than is needed for 50i or 60i SD. Those with even deeper pockets can bring footage with a measured resolution of 3.2K to the table. From any of those sources one should be able to produce boring old SD for broadcast or DVD distro that looks as good as anything shot on the most expensive SD cameras, it probably should look even better.

However the conversion process is a minefield. Following a thread elsewhere people with deep pockets have tried everything and the best solution to date is hardware scalers that cost big time. Having to feed them HD SDI and record the SD SDI output to DB puts that solution out of reach of 99% of us. The logic of buying a camera for $9K and then spending $300K for post escapes me and my bank manager :)

The people trying to crack this have tried every software solution. Nothing from the Apple camp cuts it, all the VD tricks have been tried, someone claims to have have the magic solution through AE but wants serious money to process your footage and isn't giving up the recipe any time soon. I know plenty of us here that own EX1's are grappling with this very issue. I can get a passable result but it's nowhere near as good as it should be, keeping all the resolution without creating aliasing is a problem. The real test seems to be high contrast diagonals in motion. The solution seems to lie in doing pixel tracking with adaptive filtering but so far only hardware boxes implement that in a suitable manner.

Bob.

Comments

TheHappyFriar wrote on 5/31/2008, 6:06 PM
so.... i can't just throw HD vid in vegas & render out a SD mpeg-2 for DVD with no issues? I thought I could. :/ (it has seemed to work fine for the still cam work I've been doing: 1440x1080 60i to NTSC SD DVD)
Laurence wrote on 5/31/2008, 7:25 PM
I thought most of those posts were by the FCP guys...

I think that 1440 x 1080 downrezzes perhaps a little more cleanly since 1440 to 720 is a simple divide by 2 rather than the more complex division you need from 1920 pixels. I believe that's why it has come up with the EX1 all of a sudden.
Laurence wrote on 5/31/2008, 7:28 PM
I've done some really nice downrezzes using Virtual Dub. Just set up a filter chain which separates the odd and even fields to side by side (or one over the other) images, do a lancos 4 resize, then refold the fields back together into even and odd lines.

I was doing that for a while then felt like it wasn't really that much better than what I was getting with Vegas alone. But maybe for going from the full 1920x1080 pixels it might make more of a difference.
farss wrote on 5/31/2008, 9:33 PM
Specifically this is a problem only when going from progressive HD to interlaced SD. Not specifically related to the EX1, this is an unresolved issue with the V1P (not the V1U), the EX1 and the RED. I have no doubt it's also be an issue with the F23 and F35. It's not a camera issue as such at all, it just bobs up whenever a new HD camera comes along. Gets a lot discussion that eventually dies down as we make do as best we can.
It's not a problem with telecined 35mm, they have built in magic the same as the hardware scalers.

Bob.
Laurence wrote on 5/31/2008, 10:18 PM
Why would you want to go from progressive HD to interlaced SD rather than progressive SD?
Coursedesign wrote on 5/31/2008, 10:26 PM
What is the interlaced SD for?

For ye olde authentic Cathode Ray Vacuum Tube viewing experience at The Museum of Television and Radio?

And to say that Macs couldn't do what PCs could? Very funny, since there isn't much that is available on one platform only nowadays, and what little is unique tends to be in the area of convenience and saving work hours (such as with FxPlugs in FCP).

Interlacing from progressive footage has always been difficult, even without resizing at the same time. That's just in the nature of the beast. Any movement, and you have to create fields out of thin air so to speak.

Optical flow like in FCS and PP/AE is crucial for this.

And if any interlaced-proponent TV engineers should ever pop up from their graves again, please just give them a good whack over the head with a heavy shovel so they can go back to their final rest and stay there in peace.
Laurence wrote on 5/31/2008, 10:34 PM
What about good old 30p flagged as 60i? Most episodic TV shows like "Friends" are delivered on DVD that way (as well as broadcast I believe).
Laurence wrote on 5/31/2008, 10:52 PM
Speaking of hardware scalers: don't most DVD players these days do a really good job showing 24p or 30 footage on an old interlaced TV?
farss wrote on 5/31/2008, 11:04 PM
Did I say anything could do it??
All I said was so far the ONLY solution was expensive hardware scalers, I guess which OS they have buried in their innards is kind of irrelevant.
Why go to interlaced SD. Well there's still a heck of a lot of it being broadcast and I'd say there's still more video viewed on SD CRTs than anything else. Yes, one day it'll die and I for one would cross deserts to attend that funeral too. There's many good reasons to shoot progressive, don't think I need to enumerate them, we all get it. The issue is as simple as getting the very best SD interlaced output from that. No, this isn't about trying to get the same temporal resolution from 25p as 50i, that as you've rightly noted involves inventing fields and can vary from difficult to impossible.
Clearly it can be done. I've been watching film originated video since the dawn of TV broadcast as 50i and no doubt you've been watching it in 60i with pulldown. I'd say anything going live to air these days is mostly coming from HD cameras with a mix of P and I depending on the network. The issue is getting the same results from an affordable software solution as dedicated hardware boxes. Not only is interlaced video on its last gasps but there's a bunch of new cameras that free us from the expense of VCRs and racks of hardware. The one thing missing is a top shelf HD to SD interlaced software solution. If there's one out there or some magic recipe that works I think many would be mighty glad to know about it.

Sony does have the M30 deck for SxS cards coming out shortly. It'll be interesting to see if it does 1080p to 50/60i and how it looks. Certainly one of its features is having a downscaler.

Bob.
Coursedesign wrote on 5/31/2008, 11:38 PM
Nah, Laurence got it right: 30p flagged as 60i.

Think Sony's 30PsF.

Problem solved.

And the interlacers can stay interred.

Give'm 60 lashes per second...

Goodnight.

(and G'Day to Bob).

farss wrote on 6/1/2008, 12:31 AM
Good afternoon actually and no, conversion to 25PsF or 30PsF is exactly where the problem lies.

Bob.
TheHappyFriar wrote on 6/1/2008, 5:43 AM
Might be simpler then you think: I've used the TV Out on my video card to a DV VCR & then re-captured it via my DV camera's FireWire:

here

farss wrote on 6/1/2008, 7:03 AM
That's very soft. As it's 640x480 I'd have to rescale it to feed it to a CRT which kind of negates the test. Have a look at a few frame from 25;10 onwards, I don't know what's going on there but it looks very funky, possibly resampling on top of resampling.
I'm certain it's good enough for your purposes but I know the amount of trouble games manufacturers go to make their TVCs, we supplied some of the kit to uSoft for their XBox TVCs, they did it on the cheap using the Sony 1024 Scan Converter. It looked not too hot by the time it made it to air. Other I've heard of use purpose built rigs.

Bob.
TheHappyFriar wrote on 6/1/2008, 8:58 AM
actually, it's got some motion blur because I changed the sped up the video & I wanted motion blur.

I'll do it again with some HD footage via TV out to my pyro & capture via firewire. See how it looks.

EDIT: in TV on analog I'd take high-res 3d animations & put them over the air. They looked good. Really good. I've got some HD footage I shot last night with motion (camera panning, tilting, etc) & I'll "put it through the ringer". :)
GlennChan wrote on 6/1/2008, 10:21 AM
Here is my recipe:
http://www.dvinfo.net/conf/showthread.php?t=117523&highlight=ex1

It's a hack/klduge to essentially do multi-tap filtering in Vegas... I believe that's what most of the dedicated hardware conversion devices do.

It'll be a bit slow in Vegas compared to hardware solutions... (A) because Vegas is only using the CPU instead of GPU or other more specialized hardware and (B) because Vegas is doing a lot of unnecessary processing with this method... doing GBlur this way filters pixels that aren't used.

2- An extra step on top of that is to add unsharp mask onto the SD footage to give it the appearance of looking sharper. It's a visual cheat and SD cameras do it excessively (and are limited to a radius of 1 because that's cheap/easier to implement in hardware than large radii).

3- There might be more sophisticated approaches to make things look better, e.g. adaptive techniques (handle edges differently than non-edges).