freenode :: #aomedia

16 May 2017
16:18Caspy7"AV-1" :)
18:12quikeewow VLC!
18:14quikeeAV -1 :)
18:15quikeeI decided . My next OSS will count version numbers backwards...
18:20peloverdewhen am I supposed to use xd->plane[plane].seg_dequant[seg][is_ac] vs cm->{plane}_dequant[seg][is_ac]?
18:27Caspy7quikee: so...given that I don't feel I'm distracting you from coding... :) How do you think that Videolan will contribute most to AV1? What is it's main value?
18:28peloverdeAnd why is PVQ using the quantizer with the y delta added in?
18:32smarterthe y delta can be anything but zero ?
18:32peloverdeyes, the y delta can be anythinh
18:32peloverdethere is no uv_delta
18:33smarterI see uv_dc_delta_q, uv_ac_delta_q
18:34peloverdeoops, there is on y_ac_delta_q
18:34smarterand y_dc_delta_q, but no ac equivalent
19:00quikeeCaspy7: not sure.. they are streaming oriented (VLC is a stream player that became popular as a all-around player later) so I guess it is important for them to show they are a member
19:32TD-Linuxquikee, videolan is a lot more than VLC. they also host x264, for example
19:32TD-Linuxand many of the community members are also involved in ffmpeg and libav
19:37Caspy7"The organization also supports and funds other open source projects, such as; FFmpeg, Libav, and Handbrake."
19:41JEEByes, videolan is a large umbrella org
19:41JEEBin OSS multimedia
21:49Caspy7was just reading that AV1 article and am curious...
21:49Caspy7who broached the idea of a collective to form the AOM?
21:55Caspy7more rhetorical, but wonder if Google has begun spooling up legal defenses for potential attacks :)
22:55peloverdeI think block copy is starting to look a bit silly in the analyzer
23:07TD-Linuxhow does your search even work?
23:10TD-Linuxthe result looks pretty reasonable really, given the content :)
23:10peloverdeTD-Linux: I do one "above search" in the causal area of the tile, 1 left search in the causal area of the tile. And take the best, for each search a do a 64x64(?) exhaustive mesh around the best vector.
23:10peloverde I seed both searches with the predicted vector which is the ref-mv best neighbor excluding 0,0. If 0,0 is the only neighbor I used -64,0
23:11TD-Linuxjust straight up and straight left?
23:14peloverde"above" is top left of the tile to bottom right of the previous SB row. left is top-left of tile to bottom left of previous SB in the same row. (yes there is a lot of overhead)
23:14peloverdebut most of the cpu is spent on optimize_b anyway
23:16TD-Linuxah so it's more or less an exhaustive search
23:22peloverdeit uses same motion search inter frames use but the switch to exhaustive threshold is msh lower
23:29TD-Linuxok. yeah I was thinking there wouldn't be much of a gradient to descend there...
17 May 2017
No messages
Last message: 7 days and 11 hours ago