I have a few ideas of improvements for the delogo filter included in ffmpeg's libavfilter.
As a first step I would like to clean up the code by removing one unneeded option and then removing all the subsequently dead code.
Secondly I would like to explore the possible performance gains from reordering the memory access to take benefit of prefetching. I wonder if the improvement would be measurable.
Lastly I have a few ideas which I'd like to experiment with to make the logo area interpolation closer to the original picture.
All of this will have to be done under regression testing and with performance measurements to justify the changes.
This project is one of its kind!