Jean-Marc To answer your original question, I think it's best from an implementation perspective not to impose an order on visiting blocks for deblocking if possible. It's pretty useful at least for SW optimisation to have a range of options for ordering how edges are visited and how the V and H stages interact with that.
best regards Thomas ________________________________________ From: video-codec [[email protected]] on behalf of Timothy B. Terriberry [[email protected]] Sent: 17 August 2015 21:21 To: [email protected] Subject: Re: [video-codec] Questions about Thor Jean-Marc Valin wrote: > Well, most intra predictors use only the pixels on the edges of the > block, so the interior pixels -- deblocked or not -- aren't used at all. A) That's no longer true. They run filters which may include interior pixels now. B) What I meant by "interior edges" includes pixels on the exterior edge of the block, e.g., A B *---+---*-------* | ║ | | +===+===+ | | ║ | | *---+---*-------* Here ║ and = represent interior edges which need to be deblocked, and the four blocks in A are used to predict the single block in B (though it would not matter how B is split). The interior edge separating the top and bottom blocks includes pixels on the very right edge of A, and it matters whether they are deblocked or not before predicting B. _______________________________________________ video-codec mailing list [email protected] https://www.ietf.org/mailman/listinfo/video-codec _______________________________________________ video-codec mailing list [email protected] https://www.ietf.org/mailman/listinfo/video-codec
