DGAVCDec 1080i mbaff and deinterlacing.

[mitsubishi]

Hi, thanks a lot for everyone involved with this project, it's been a few months since I last did AVC 1080i to AVC 720p encoding and with the updated x264 and this source filter its absolutely flying now (relatively speaking)

I still find interlacing a confusing issue at best, it's obvious when there is the combing. With mpeg2 megui will analyse the source and I just trust it and the results are great. I compared a fielddeinterlace() which staxrip uses to the (slow) one it suggested and it removes the blurring and brings out detail beautifully.

Anyway I asked this before when using CoreAVC as a source filter through DS and was reassured that it comes out deinterlaced. So is this still the case, I can take the "AVCSource" and not have to do any deinterlacing? It's just motion can be a little blurry, it's probably supposed to be, but I wanted to check.

I'll go snag a screenshot.

[Guest]

Nope, there is no deinterlacing done by DGAVCDec. If you want it, add a deinterlacing filter in your Avisynth script.

[mitsubishi]

Oh right.

From what I gather MBAFF means it can interlace part of the screen when it feels like it. For what it's worth here are a couple of screen shots. The blurring is present whenever there is fast movement, but the combing only seems to be there on the wipes, which made me think it was the intended style.
http://www.mytempdir.com/1324738

This is kind of why I thought the decoder did the deinterlacing, because it knows when it should. I did run it through h264_parse, but have very little idea what it all means.

Any hints on how I should be deinterlacing?

[Guest]

Hard to say why you have progressive motion blur on some shots and combing on others. Maybe one of their cameras for some angles is a high speed one that shoots progressive.

The reference decoder does not deinterlace so it doesn't use the MBAFF indication for that. It's an interesting idea to use that information to guide deinterlacing, but remember that an area could be combed and yet be coded progressively; it all depends on how the encoder works and what it decides uses fewer bits or renders the frame better. It would be worth doing some experiments in that regard. Maybe one could apply a less aggressive deinterlacing in the progressive coded areas, giving a better overall fidelity.

There are many deinterlacers and only you can pick the tradeoff of artifacts vs. performance that is right for you. You could look at these to get started:

TDeint
LeakKernelDeint
FieldDeinterlace(blend=false,full=false)
TomsMoComp

Please, no what's the best deinterlacer followups! :)

[mitsubishi]

I thought the BBC only used Panasonic cameras, but after rechecking that was only one division (BBC factual). This particular show is co-produced with an American company, but I doubt AMC have provided cameras.

I did think that as it only combs on the wipes that the interlacing was introduced in the edit, they use final cut pro afaik. But not knowing much about how it all works, it's just uninformed speculation.

I'll get some reading done over the weekend and play around a bit.

I did last night blindly copy this out of another meGUI generated script and that removed the combing, but I don't know kind of damage it was causing.

edeintted = last.AssumeTFF().SeparateFields().SelectEven().EEDI2(field=-1)
TDeint(order=1,full=false,edeint=edeintted)