Question about interlaced MPEG2

[Entropy512]

I have a question about interlaced MPEG2 that I haven't been able to figure out the answer to even after making an attempt at reading the sources to Xine's MPEG decoder:

How independent are the fields? Is it possible to create or modify a decoder so that either the even or odd fields are simply ignored? Or is there sometimes dependence between even and odd fields?

This would significantly increase decoding speed in cases where the desired output resolution is less than half that of the input. I know trbarry has a nice AVISynth plugin that does this after decoding (I forget the name at the moment, but it's good for resizing 1080i down to 960x540 very quickly.), but from reading one of the recent MPEG2DEC3 threads, the decoder itself spends a decent amount of time internally just copying data - It could be sped up significantly if it were able to drop even or odd fields.

It would probably increase decoding times even if decimation were to occur after decoding but before passing the data on to AVIsynth, so that AVIsynth would never have to deal with the data. (One less copy of a full raw field).

[trbarry]

It might be a bit faster but not much and maybe not any. In my YV12InterlacedReduceBy2 (or whatever I called it) I can't just select even lines of chroma data because of the mutated way chroma is stored in MPEG-2. So I have to interpolate between 2 lines to get the right answer.

If you just returned the even lines it would not match any known standard and would have a small vertical chroma delay if you just treated it as progressive YV12.

- Tom

[Entropy512]

OK, thanks for the info.

The impression I got from the xine sources was that there was indeed some interdependence between the even and odd fields, but I figured it was best to see if one of the experts here had some insight on the issue.

Oh well.