I guess, it is because of dependency between pixels at decoding. What if to group tiles with the same predictor type?
I said "guess" because I'm unable to locate the code for encoding and decoding -e 1, the structure of libjxl is a bit over-complicated. Can you provide the direct links at current master?
Or, maybe, you want to test this idea and write fjxld?
Hi, Luca.
The question is related to jxl.
Previously I noticed that decoder is 10 times slower than encoder: https://github.com/libjxl/libjxl/pull/1975#issuecomment-1416775973
I guess, it is because of dependency between pixels at decoding. What if to group tiles with the same predictor type?
I said "guess" because I'm unable to locate the code for encoding and decoding
-e 1
, the structure oflibjxl
is a bit over-complicated. Can you provide the direct links at currentmaster
?Or, maybe, you want to test this idea and write
fjxld
?