direct_pred=<name> [spatial]　ageha was here

Determines the type of motion prediction used for direct mac-
roblocks in B-frames.
none Direct macroblocks are not used.
spatial Motion vectors are extrapolated from neighboring blocks.(default)
temporal Motion vectors are interpolated from the following P-frame.
auto The codec selects between spatial and temporal for each frame.
Spatial and temporal are approximately the same speed and PSNR,
the choice between them depends on the video content. Auto is
slightly better, but slower. Auto is most effective when com-
bined with multipass. direct_pred=none is both slower and lower
quality.

x264 [info]: direct mvs spatial:87.8% temporal:12.2%

ああ

• temporalはゴーストが出るとの情報有り。
• Doom9コデックコンテスト2005では決勝の2例(エンコードがより困難な素材)で空間軸(spatial)を使っている。
• diret_pred=autoの登場により、ブロック毎に適切な方式を使い分けられるようになった。spatialとtemporalの比は概ねアニメで9:1。実写で7:3程度。