Spatial scalability (SVC)
One AV1 or VP9 stream carrying a lower-resolution base layer and the 720p picture, so a receiver
or relay can drop to the base without a second encode (--scalability L2T1 --preset medium or fast,
and for three layers --scalability L3T1). Rate is the total stream rate;
quality is the 720p output, against single-layer x264 medium (default tuning).
A small development set. These are development clips, not held-out ones. The libaom and
libvpx rows use those projects' example SVC encoders (one thread, CBR, no frame drops), which demonstrate the
API and are not tuned products; other settings may do better. Speed is partly
estimated: on the PC most points were timed in pairs with x264 medium, the others are scaled by instruction
counts, and the Jetson column applies the PC's ratios to the Jetson's timed fps.
Settings, the K-SVC modes and VP9 three-layer results.
Bitrate saving of the 720p output against encode speed
BD-J at the total rate, vs single-layer x264 medium · lower is fewer bits, right is faster