Film source: IVTC of course; but deinterlace as well?

[McQuaid]

First off, I capture at 352x480 for full frame. I frame serve from virtual dub to tmpgenc. I use tmpgenc's automatic ivtc, and I notice it has additional options for deinterlacing. Since I am encoding to vcd film, should I deinterlace in addition to ivtc? And if so, which one de-interlace method is best under tmpgenc? And if i don't de-interlace what does tmpgenc do by default with an full frame source? Just squash it down and drop one field?

[VidGuy]

If the IVTC is good (and I've only rarely had a problem with TMPGEnc's automatic IVTC), there's no reason to deinterlace - it just loses resolution. However, since you're making a VCD, presumably by dropping every other scan line - you get automatic de-interlacing, so between the IVTC and the automatic de-interlacing of dropped scan lines you get a very good picture......

However, I'm not sure how TMPGEnc converts from 480 to 240 lines - I use AVISyth's ReduceVerticalBy2 - that way, I know it's dropping scan lines and not doing some odd bilinear or bicubic resizing algorithm.....

[Vejita-sama]

Ways to deinterlace:

1) Convert to ???x240 doesn't really matter how you do this.
2) If telecined source, force film in DVD2AVI. This is NOT the same as IVTC.
3) IVTC

If you'r capture is a 29.97fps true interlace source (ie. not telecined) you should either de-interlace it, resize to ???x240, or encode to MPEG2 which supports interlacing.

[stanwebber]

avisynth's reduceby2 is not near identical, but exactly identical to virtualdub's 2:1 reduction (high quality) which is a 3x3 kernal bilinear resize filter. in other words you are blending fields, not discarding them. SeperateFields plus SelectOdd or SelectEven will discard a single field. add AssumeFrameBased plus ComplementParity if needed to nail down the field-order. unless things have changed since v12a(i know it's been a while), tmpgenc uses a resizing algorithmn rather than discarding a field. deinterlacing after auto-ivtc is not ill-advised, but none of the deinterlacing methods in tmpgenc(again v12a) are suitable for this. use an area-based deinterlace filter set to a low threshhold to clean up the misjudged frames while leaving good areas relatively untouched. i know donald graft and gunner thalin have filters that work along these lines. check the virtualdub homepage for links

[McQuaid]

Thanks for all the help.

Since I'm frame serving to Tmpgenc (to avoid resaving due to additional drive space needed and to avoid resaving with picvideo at 19 setting which would encur more picture quality loss) I'd prefer if there was a method to deinterlace within tmpgenc that didn't just discard fields, as I assume a blend of the full frame would retain more quality.

But if not I guess I'll have to use virtualdub.