[Sls-mhdc] fast neighbor prediction mode

[Enrico Magli]

Dear all,

I have uploaded in CWE private --> 123-1.B two files describing experiments for a fast predictor in full neighbor oriented mode.

File fast_neighbor.pdf describes four possible configurations of the predictor, while file fast_neighbor.xlsx reports results in comparison with the fully-fledged fast neighbor mode. Two configurations, FF1 and FF4, seem to work reasonably well. Generally speaking, their performance is better than the corresponding reduced column-oriented predictor, though there is indeed some performance loss with respect to the full neighbor oriented mode. Note that bit-rates have been generated using the range encoder, so only bit-rate differences are significant.

I guess which of these or other configurations to possibly choose will also depend on Lucana's analysis of implementation issues.

Regards,
Enrico