You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We currently retain the first/last keyframe of every segment, but this is overkill. We could retain first/last of clip, and allow last of segment to be stripped. When we decompress, if the second sample isn't found within the expected segment, we know it is the first sample of the next segment. We can extend this to allow stripping of the first keyframe in a segment as well. We simply need to scan our segments headers forward/backwards. This might be slightly slower during decompression but in practice the segment header metadata touched likely lives in the same cache line as the expected segments.
It is unclear if this would be a win. In practice, we would only really benefit if we can strip whole segments but that would only happen if a character is stationary for some time. This is more likely to happen in a cinematic but not in gameplay animations.
We have on average 16 samples per segment. If we freeze the first/last keyframe, we can only strip 87.5% of keyframes. Each keyframe contributes 6.25% to the footprint. If we can strip a whole segment, we can also strip its metadata which is considerable.
The text was updated successfully, but these errors were encountered:
We currently retain the first/last keyframe of every segment, but this is overkill. We could retain first/last of clip, and allow last of segment to be stripped. When we decompress, if the second sample isn't found within the expected segment, we know it is the first sample of the next segment. We can extend this to allow stripping of the first keyframe in a segment as well. We simply need to scan our segments headers forward/backwards. This might be slightly slower during decompression but in practice the segment header metadata touched likely lives in the same cache line as the expected segments.
It is unclear if this would be a win. In practice, we would only really benefit if we can strip whole segments but that would only happen if a character is stationary for some time. This is more likely to happen in a cinematic but not in gameplay animations.
We have on average 16 samples per segment. If we freeze the first/last keyframe, we can only strip 87.5% of keyframes. Each keyframe contributes 6.25% to the footprint. If we can strip a whole segment, we can also strip its metadata which is considerable.
The text was updated successfully, but these errors were encountered: