A complex chain of filter interaction. Possible?

[Die*wrek*show]

I had an idea that I'd like to share, I doubt it's been discussed before in fact it might seem crazy.

I was looking on the filter page, and noticed how many avisynth filters there are, and how many possiblities with all those variables to be set in each filter.

So my question-

Would it be possible, to have an automated filtering process, that makes decisions based partly on analysing the video and partly on your input? Or maybe IF THEN type things?

Like, for instance. A front end program, opens a video and tries a combination of filters(or filtering processes) then posts a clip, asking you which one looks best so far. Then it continues using a diff combination of filters, then asks again, ect... so it tries an array of avisynth scripts seperately, to try to find the right clip that you say looks best. now all the clips might look the same, so the clips would have to be magnified or otherwise manipulated to bring out the defects for previewing.

I've been around for a while but all i use mostly is donald graft's dgmpgdec, and virtualdub. so i don't know that's why i'm asking, if it already exists then sorry.


edit: my mistake, donald. i was lazy with my typing. i meant to say "donald graft's version of the program that started as dvd2avi" but didn't feel like typing all that. i still think of it as dvd2avi, force of habit i guess.

[Dark Shikari]

That actually wouldn't be a bad idea, but it runs into two problems:

1. There are temporal filters; its not necessarily fair to judge based on a single frame. This probably isn't too big a deal though.

2. This only works if you're dealing with a very simple linear filter chain. Some of the more complex scripts cannot be automatically created in such a manner.

[Die*wrek*show]

I don't laugh very often on the doom forum, but tonight I did.

In my original post, I forgot to add the link to the thread that got me thinking about this, I was really impressed by this guy and his script to clean up youtube videos. So then I found it (http://forum.doom9.org/showthread.php?t=134078) just now, and guess who orignally posted it? YOU


what are the chances? :lol: talk about synchronicity.

[Die*wrek*show]

well as far as your responses go, i'll try to respond to #2 as best i can and it SHOULD kinda answer #1

What I had in mind, was a bunch of templates with the paths and variables that start out mostly filled, but then get progressively less filled as the "filter engine" runs and gets closer to the point in needing user interaction.

of course this all assumes there exists some master table of reference to find the most effiecient way to handle a certain type of video. like, the most ideal way to handle a crappy interlaced 25 fps source are xyz filters, with x to do so and so, y to do so and so, ect.... without that of course the individual filter engines couldn't be built to produce the most ideal results in terms of clean video, space efficient, no artifacts...

the engine would actually be a bunch of smaller engines that would be suited for a certain type of video, which the "analyze" phase would have already determined.

[foxyshadis]

avsp includes a neural network function that modifies filter chains to get as close as possible to some desired output, but that wouldn't really work if you don't already have the output! MeGUI's interlace mode detection could be rewritten for avsp fairly easily, I think, and perhaps other things could be tested for if anyone has any really good ideas. But interlace types was hard enough to get nailed down, and that's painfully obvious to the human eye.

Some people have posted big scripts that you can just check off the parts you need to enable. Like "interlaced, field blended, sharpen, resize". I think they were in the avsp thread.