Richard Berg
12th December 2002, 14:05
Some friends of mine are trying to IVTC the Star Wars laserdisks. They're usually better than I am when it comes to wringing the most out of video-ripping tools, and sure enough all I could do was whip up some tools to facilitate the creation of override files.
However, I knew Donald would be interested in finding more "pathological" cases to test future refinements against, so here goes:
This segment (http://freelancepolice.com/temp/testhuff02.avi) (44MB Huff) cannot be IVTC'd by any of the automatic settings we've tried (pay attention, e.g., to the blockiness of the little planet). As a quick example, this brief section (http://www.freelancepolice.com/images/swtest03.avi) (100KB) shows the results of manual intervention and then letting it off on it's own halfway through.
What we've theorized is that very dark, nearly empty frames don't provide enough information for the field-matching algorithms to do their thing -- even the very high motion scenes you'd think would cause problems have fared quite well, so long as they have enough contrast.
This has been more or less confirmed through DebugView: when the troublesome clips above were first passed through a crazy filter like Tweak(sat=0,cont=11), the result looked terrible of course, but Decomb was making the right decisions. This fact immediately lends itself to a 2-pass method that transforms the DebugView output into override files appropriate for the untweaked movie, but naturally if Decomb itself can be improved then so much the better. I'm not sure how adding a rule checking for low-luma scenes might affect general operation, but it's something to play with at least :)
Edit: ha! wotef cross-posted...
However, I knew Donald would be interested in finding more "pathological" cases to test future refinements against, so here goes:
This segment (http://freelancepolice.com/temp/testhuff02.avi) (44MB Huff) cannot be IVTC'd by any of the automatic settings we've tried (pay attention, e.g., to the blockiness of the little planet). As a quick example, this brief section (http://www.freelancepolice.com/images/swtest03.avi) (100KB) shows the results of manual intervention and then letting it off on it's own halfway through.
What we've theorized is that very dark, nearly empty frames don't provide enough information for the field-matching algorithms to do their thing -- even the very high motion scenes you'd think would cause problems have fared quite well, so long as they have enough contrast.
This has been more or less confirmed through DebugView: when the troublesome clips above were first passed through a crazy filter like Tweak(sat=0,cont=11), the result looked terrible of course, but Decomb was making the right decisions. This fact immediately lends itself to a 2-pass method that transforms the DebugView output into override files appropriate for the untweaked movie, but naturally if Decomb itself can be improved then so much the better. I'm not sure how adding a rule checking for low-luma scenes might affect general operation, but it's something to play with at least :)
Edit: ha! wotef cross-posted...