New hybrid clip handling

[Guest]

I've just implemented a new way of rendering hybrid clips. It is described in my website journal.

I'd appreciate your opinions about it.

[Richard Berg]

As a sci-fi fan, I like it! For comparison, can you encode a clip using the "MotionPerfect" technique you used in a previous thread?

Nota bene: Divx 5.02 stopped decoding the video when it hit the scene change (c. frame 105 in the decimated vids). Downloading version 5.05 fixed it.

[sh0dan]

I've been doing similar stuff using smoothdeinterlace(doublerate=true) with excellent results - by far the best hybrid conversion I can think of.

The only thing I have is, that if you use pure bobbing, is that it has to be intelligent (as smoothdeinterlace), to avoid edge blending/blurring, but I assume that's what you're doing.

[Richard Berg]

I'm not sure that's a fair comparison. The goal of this new mode, as I understand it, is to fit 30fps video into 24fps progressive with minimal jerking; doubling up to 60fps looks nice but defeats the purpose.

Also, does that mean you're bobbing entire hybrid clips? I can't imagine that working very well -- even though PAL doesn't have the weird field matching of NTSC telecine, you're still separating fields temporally that were originally together (on film).

[Guest]

Originally posted by sh0dan
The only thing I have is, that if you use pure bobbing, is that it has to be intelligent (as smoothdeinterlace), to avoid edge blending/blurring, but I assume that's what you're doing. Very perceptive of you, Mr. sh0dan. Best results are indeed achieved using the equivalent of a "smart-bobbing" approach on a per-cycle basis.

[Guest]

Originally posted by Richard Berg
I'm not sure that's a fair comparison. The goal of this new mode, as I understand it, is to fit 30fps video into 24fps progressive with minimal jerking; doubling up to 60fps looks nice but defeats the purpose.

Also, does that mean you're bobbing entire hybrid clips? I can't imagine that working very well -- even though PAL doesn't have the weird field matching of NTSC telecine, you're still separating fields temporally that were originally together (on film). Your point about fairness is of course valid. But sh0dan is right about the similarity of the ideas, because you can take sh0dan's output and do this with it to get effectively the same result as mine:

SelectEvery(10,0,3,5,8)

But using that exact approach, i.e., smart bobbing to 60fps and then doing SelectEvery is not a practical approach for a filter that is handed a cycle of five frames. As you correctly point out, the decimator has to apply it only to the video sections.

[Guest]

Originally posted by Richard Berg
As a sci-fi fan, I like it! For comparison, can you encode a clip using the "MotionPerfect" technique you used in a previous thread? See here:

http://shelob.mordor.net/dgraft/journal/fox24-motionperfect.avi

It looks fine, if you can overlook frame number 106! :):):)

[sh0dan]

>It looks fine, if you can overlook frame number 106!

Wow - this is a perfect example of motion estimation miserable failing! IMO this single frame just proves that for this job smartbobbing is the only clean solution for decimation.

How is the top-edge in the uncompressed sources - it seems like it's either being very hard compressed - or there are some of the typical marching ants problems. Even though I assume it's a compression problem.

[Richard Berg]

Do you have any idea what kind of algorithm it uses? "Secret" looks pretty good, but the motionperfect clip I had to open in Vdub to verify it was actually 24fps :cool:

It looks fine, if you can overlook frame number 106
Dunno how you noticed this, but it made me laugh! Like warpsharp gone horribly wrong. I suppose it shows that there's some attempted motion compensation in the mix...

Edit: sh0dan posted while I was typing. Not sure if I agree -- the single error frame was unnoticeable in full-speed viewing, and regardless the smoothness more than made up for it.

[mf]

Originally posted by sh0dan
How is the top-edge in the uncompressed sources - it seems like it's either being very hard compressed - or there are some of the typical marching ants problems. Even though I assume it's a compression problem.
To me it looks like a typical interlacing problem on black borders.

neuron: your new method definitely looks better :). I'd positively vote for it in Decomb.

[OUTPinged_]

The video clip Mr.Graft put to his site looks excellent, but i wonder how it will look on animation hybrid clips or any other clean material. Did anyone tried it already?

[edit=typos]

[Defiler]

Originally posted by Richard Berg
Not sure if I agree -- the single error frame was unnoticeable in full-speed viewing, and regardless the smoothness more than made up for it. I haven't stepped through any of these clips.. just watched them at normal speed..and I had no trouble seeing the error the first time. It wouldn't take many of those to make me unhappy. Heh.

I'm very glad to see progress being made on hybrid clips, though. Mode=Grail looks like a big improvement. Heh.

[MasterYoshidino]

yep really need to try it on those animations (oh yeah chobits 1 had a hybrid scene at the preview when Chii was shown with the clothing on, camera pans from bottom to top of screen, filmed on video at 30fps ... on a 24fps series >_<)

[mf]

Chobits IS hybrid. It's all 24fps animation with 30fps tweening. And the fansubs (and thus the raws) were totally crappy in that. Worst jerking ever >.<.