Torgeir Veimo (torgeir@xxxxxxxxxx): > > Unfortunately, my algorithm is two-pass, so it is non-realtime. As > > such they wouldn't be interested, but VirtualDub would (I emailed that > > guy but he hasn't got back to me, plus, I just 'finished' my alg a few > > days ago). > > > > I have a realtime algorithm I use in my DVD player (movietime, > > www.sf.net/projects/movietime), but I haven't compared results > > with dscaler's. > > How many frames do you need? I think descaler buffers as many frames > as the algorithm needs. Or is the processing power too high for > realtime? Well, I get a bunch of very slow-to-compute stats on every frame in the video and write them in an index. Then I figure out my best-guess pattern for every 5 frames (so, need a buffer size so far of 5). Next, I go through that pattern and watch for changes, then I decide how best to handle that change based on the stats of the frames in that 10-frame sequence. So, I need about 10 frames, and currently the CPU usage is very high. You can adapt my algorithm so it doesn't handle changes as nicely, and so it only requires a buffer of 2 frames, but then I bet it would be equivalent to dscaler's algorithm. > It would be nice to standardize on a module api for deinterlacers > under linux, both for pctv deinterlacers and dvd deinterlacers. Yeah, I was working on a 'libvideo' library with a bunch of these useful things a few months back, but I didn't finish. I was interested in realtime algorithms there for my DVD player (www.sf.net/projects/movietime). I do intend to finish this work, probably doing a lib of useful stuffs with omega of gstreamer.sf.net. #gstreamer/#livid on irc.openprojects.net has been cool for discussing video hacking in general. -- Billy Biggs vektor@xxxxxxxxxxxx
