Curiosity is killing this cat...

Spot|DSE wrote on 12/18/2004, 6:13 PM
Playing with some m2t media, putting it on an HD timeline w/in Vegas, and matching DV. I've got a scene that was shot in DV and also in HDV at the same approximate time for purposes of comparison, shot with the same camera.

Punching up colors and saturation in the DV file, and using the m2t as a baseline after converting the m2t to an avi using CineForm's intermediary, it's pretty shocking how well Vegas can upsample the DV to *almost* match the HD. What's amazing, is that the macroblocking is not visible, except in the extremely high motion footage, and that's to be expected with this sort of format anyway.

BJ_M or anyone else, any comments on why you think this might be? I've been in many situations where VHS looks better after transcoding, and I understand the reasons why that happens, but can't quite put a finger yet on why Vegas is doing the upsampling job it is, when we're going 4 times the size. I'm changing the PAR when upsampling, as that's a requirement I figured out doing the QT upsamples.

BTW, this is a VERY slow process. 10 secs of DV takes 3 minutes to render to a TS stream. Ouch.

Comments

farss wrote on 12/18/2004, 7:57 PM
I'll hazard one guess, the encoder can take it's time unlike the one in the camera. Also as you're coming from a lower res source (i.e. less detail) there's less for the encoder to cope with.
As to the perceived resolution, I think firstly you need to see it on a BIG screen. From what I've seen cutting SD (as in 4:2:2) up against HDV, the SD look horrible on a big screen. Yet I've projected heaps of DV25 and it doesn't look that bad. Curious?
I think this has a lot to do with the upsampling interpolation. When you see the DV25, the brain maybe only expects to see so much, the very visible scan lines give a visiual cue. When it's upsampled the cue is gone so you just see yuck.

The trend down here for transmission will be for a long time to do just that, upscale SD in a Terranex and broadcast that as HiDef. It save a huge amount on infastructure costs. It's not the cost of the decks, it's the routers, tie lines, the whole station / network may need a rebuild. Doing the upscaling at the transmiiter save a bundle.
Bob.
Spot|DSE wrote on 12/18/2004, 8:30 PM
I'm viewing it on an 11' screen, via an XGA projector. I don't have a true HD projector available to me right now, but even at 1024, I've usually been able to pick out the garbage. I'm just not seeing it. Granted, it's not as clean as the HD is, but it's still very surprising. I've asked for clearance, I'll post stills if permissible.
I'm wondering if in the render back to the m2t that it's doing some smoothing based on the format.
taliesin wrote on 12/19/2004, 3:29 AM
I once tried to help somebody who needed to have a film transfere of a dv production shot whith the Pana DVX-100. He used a rather time consuming workflow of exporting image sequences out of Vegas to use them in a certain image processing software. There he made the upscaling and sharpening process. His results looked great!

Now I was wondering whether this workflow could be improved - doing all inside Vegas. But without loosing quality compared to the process he took. The guy sent me some of his DVX-100 shots and I started testing.

In the end I simply set the Vegas project properties to the hi-res format, I improved the levels a bit - especially the black levels, I applied a smart smoothing to reduce the enlarged video's noise and combined this (using a composition) with the Unsharp Mask function to sharpen the edges a bit. All the values used there were rather tiny. The danger is to overdue it here.
It was rendered to the hi res format and finally the result looked even better then what my friend tried outside of Vegas. Long render times, of course. But I was so amazed of the output. I would not have believed upscaled DV can look that good.

When I got the first HDV shots and comparable DV shots I tried this once again and was happy to see what a great job Vegas does in upsampling Video. It's a pleasure to have a tool like Vegas by my side! :-)

Marco
SimonW wrote on 12/19/2004, 7:06 AM
This is also why I am desperate to see some upsampled DV using the Algolith CAS plugin should anyone have it (their demo still costs $25!!) The examples on their site look extremely impressive with upsamples that retain sharpness as well as fully antialiasing sharp edges.

It's just that I'm not prepared to spend $25 to try it out for a week only to find it's not as good as they say.
Spot|DSE wrote on 12/19/2004, 7:32 AM
I've written to Algolith a couple of times to ask if they'd like me to review it for a magazine or for the DMN. They responded with "Sure! Just buy the demo"
My thought was, "I wonder how fast you're working to put yourselves out of business......" Charging for a demo isn't totally unheard of. Wanting a journalist to pay for a demo is indeed unheard of. That's just plain stupid.
Taliesin, I've done a lot of upsampling, but this is the first time I've attempted to match footage shot on the same camera, simply because you can't find cameras that shoot DV and HD through the same lens. That's where the shocker comes in.
Between Gaussian blur, unsharpmask, Black Restore, and Mike Crash's plugins, you can get reasonably close to what you're seeing in HD from the FX1/Z1 cameras.
The main reason I'm doing this, BTW, is because I've got a MONSTER stock footage library that's all in SD, and I'd really like to find a means of not having to sell it off.