Can x264 encode slice by slice?

[chen]

H.264 Standards indicate that a picture can be or partitioned into more than one slice or encoded as a whole ( in this case the pic is considered as a single slice). But is x264 able to encode a frame slice by slice? If so, is it automatically decided by the encoder itself how to partition a pic into slices, or is there a parameter that users can set in order to affect the way of partitioning?

[stephanV]

slicing is used if you use multithreaded encoding AFAIK.

[chen]

oh, i see. The parameter --threads <integer> enables paralell encoding, whichs means slices will be used. But is a frame to be uniformly partitioned into slices?

[stephanV]

I think yes. I doubt sliced encoding could give you any improvement in quality if that is what you are looking for. So considering that and the reason why slices are used (to speed up encoding) it probably makes most sense to have equal sized slices (if anything else is actually possible)

[708145]

I think yes. I doubt sliced encoding could give you any improvement in quality if that is what you are looking for. So considering that and the reason why slices are used (to speed up encoding) it probably makes most sense to have equal sized slices (if anything else is actually possible)

slices decrease quality at same bitrate.

[Kopernikus]

IIRC x264 partitions the picture in macroblock raster scan, so with 2 slices the fist half of rasterscanned macroblocks belong to slice 1 and second half belongs to slice 2.

[omion]

IIRC x264 partitions the picture in macroblock raster scan, so with 2 slices the fist half of rasterscanned macroblocks belong to slice 1 and second half belongs to slice 2.
I've heard that too, but I've been wondering: why do they do it that way?

From what I know, the only reason that quality decreases is that you can't do spatial prediction between the slices, and therefore it would make sense to make the "stitch" as small as possible. Since most movies are wider than they are tall, it would make more sense to use the left/right halves.

(Generalizing, you want all the slices to be as close to a square as possible, because that maximizes the number of internal macroblocks)

... but that's assuming I know what I'm talking about :D

[chen]

The parameter "--threads " enables us to partition a frame into slices and each slice is packaged in a NAL unit which is called "coded slice NAL unit". According to H.264 syntax, a coded slice NAL unit of a non-IDR picture can be further partitioned into 3 NAL units, and this is the so-called "slice data partitioning". I wonder whether x264 enables this function.

[Kopernikus]

x264 doesnt support slice data partitioning

[Sirber]

slices decrease quality at same bitrate.I heard it was about 0.1db.

[Sharktooth]

It can be even less (or maybe more). It depends on the source... however as a general rule less slices = more quality.

[puffpio]

what about an encoder farm thing

like 1st pass the master computer determines where all the I P B frames are....
then for the 2nd pass it assigns cpu1 to frames 1-10, cpu2 11-20...etc...and then combines it all together?

the way teh work could be dividided up is a chunk of frames..beginning w/ an I frame...that way each chunk of work is independant of the other...
once the encoding of a chunk is done it could report back to the master process about rate information and stuff...
could work for multicore, multiprocessor..and even a clustered network situation

[akupenguin]

It's being done: ELDER (http://forum.doom9.org/showthread.php?t=87715&highlight=elder)