Comments

Spot|DSE wrote on 7/15/2004, 9:14 PM
Why not try rendering a short segment in both formats, and you'll see why...
Phil_A wrote on 7/16/2004, 6:38 PM
My tests have been inconclusive. It seems to be a crap-shoot, which method will yield the best results for any given shot.
Spot|DSE wrote on 7/16/2004, 6:46 PM
Try it with test media, or with fine details that are both horizontal and vertical.
It isn't always visible with normal media. But you'll find, and Sony along with others, recommend blending. You'll see the diff if you use fine detail.
Are you viewing on an external monitor or computer monitor?
Phil_A wrote on 7/17/2004, 2:07 AM
I can definately see what's going on, with the manipulation of the fields... But the stuff I'm working with has a lot of motion (which may be the most significant variable) and what I'm seeing is:

On one shot, blending the fields yields the most pleasing results, yet on a another shot interpolating looks much better.

Sony suggests using 'blend' for better detail and 'interpolate' to better handle motion. My tests would suggest that the prefered method is going to depend on a lot of factors with the composition of a shot... And (based on my tests) I can't figure a way to predict this.

Now the reason I asked for your thoughts, is because, in your article, you specifically recommended blending. And it seems to me that your reason might be because of the finer detail... But I'm discovering that different shots may require different treatments and there probably isn't an axiom for this type of processing.

My tests have been viewed on both external and computer monitors, though the result on the external monitor is certainly more significant... Although I have enjoyed stepping thru each frame (on the computer monitor) to more closely examine the price of changing the frame rate and deinterlacing.

(Thanks for helping out as I explore this!)
Spot|DSE wrote on 7/17/2004, 5:37 AM
You're right, there is no axiom. (is there in anything creative? :-))
However, Blending vs Interpolating *generally* looks better in keeping the softer look and keeping detail. But your mileage may vary.
There was a hellacious thread on this in the FCP and Filmlook forums on the DVinfo.net, and even the FCP cats were raving about how great Vegas does this, and why can't any of their tools on the Mac work this well, etc. The consensus from everyone, regardless of their platform, was that Vegas does this better, and that on the whole, they liked Blend best.
farss wrote on 7/17/2004, 6:25 AM
Phil,
what you are seeing and what SPOT is saying both make sense to me. You've got three choices whne it comes to de-interlacing:
a) Throw one field away, interpolate the missing lines from what left and duplicate that into the missing field. You throw away half the vertical res doing this.
b) Blend both fields. Works great until youve got a motion, for the things that are moving you get two images of them on alternate lines in different positions, some blurring will merge them but again with a serious loss of horizontal resolution.
c) Use b) with motion compensation. Best solution but needs expensive software / hardware.
Now what POT and Sony are saying makes sense if you shot the video like you would film, keep the camera still and let the action move in the frame. Blending will make the moving objects blur but we expect that, the rest of the frame will retain resolution. Alternatively have the camera track the action so it stays still in the frame. In this case again with blending the moving bits blur but we expect that.
Where I think it'd come unstuck big time is say panning one way with the action moving the other or mistracking the action, everything has to become blurred. Our eyes will forgive part of the frame being blurred, when nothing is clear we really notice it.
That said remember this at best isn't a perfect process, you maybe looking too critically at the results. Get someone who doesn't know what you've done to give you an A/B evaluation.

Bob.