Repository navigation
Optimize memory consumption of ResizeProcessor #642
Description
Activity
@antonfirsov I'm assigning this one to you for now since you're more likely to get it right than me. Hoping someone from the community will step up though and give us a breather.
Reacted by Anton Firszov
Currently we are using
O(NewWidth * SourceHeight)memory for doing resize operations, however it's possible to do it consuming onlyO(NewWidth * KernelDiameter)memory, without significant negative effect on speed.We need to implement a "sliding window" algorithm for this.
It will most likely fix issues like #597.
UPDATE
The issue is now part of the epic #733.UPDATE 2
The ideas in #733 are abandoned now, we need to implement this with the current architecture.