On Fri, Nov 20, 2015 at 10:33:04AM +0100, Clément Bœsch wrote: [...] > > i think this doesnt support libavcodec/tpeldsp.c as used in svq3 > > that would need a motion_xy / 3 > > > > i might be missing something but > > > > * Motion vector > > * src_x = dst_x + motion_x / motion_scale > > * src_y = dst_y + motion_y / motion_scale > > */ > > int32_t motion_x, motion_y; > > uint8_t motion_scale; // This is almost always a power of 2 > > > > should be enough to support all things > > > > ok, so always a div then > > > The "*motion_scale" should not be needed, the exportet vectors > > can just be multiplied up if thats needed > > what is the range of mv_scale in snow? how much precision i'm going to > loose by doing avmv->motion_scale = mv_scale/8? >
ah wait my bad, forget this. note: can't we make the motion_scale always a power of two by pre-multiplying motion_[xy]? -- Clément B.
signature.asc
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel