Quantizer <-> Quantization matrix tradeoff

[15081947]

This is my understanding.In Xvid encoder,
1. Within the targetted filesize, the best quality is obtained by using quantisers 1 or 2 as much as possible
2. The best quantization matrices (HVS-best,...) end up using higher quantisers.

Given that I can fit the rip in the targetted size,
What is the operating point? H263 with low average quantizer size OR HVS-best with high average quantizer?

[pest]

the used quants don't tell you much about quality.
you can only compare quants with the same quant-matrix.
so quant2 with h263 looks better than quant8 with h263.
choose the matrix of your choice and look at which quants
the artifacts start to annoy you and encode to that target.

[Teegedeck]

This is my understanding.In Xvid encoder,
1. Within the targetted filesize, the best quality is obtained by using quantisers 1 or 2 as much as possibleDon't ever use quantizer 1. It looks just like '2' but is a whole lot more wasteful.
2. The best quantization matrices (HVS-best,...) end up using higher quantisers.Not necessarily. If you define 'best' as 'most efficient' then the anwer would be 'no'. If you define 'best' as 'best quality regardless of filesize' then the answer would be 'yes' (if you aim for the same filesize with h.263 and SixOfNine, then the quantizers of SixOfNine will be higher). But such matrices were really tailored for higher quality at bigger filesize than H.263 at quant=2. Because, as said, quant=1 is nonsensical.

Given that I can fit the rip in the targetted size,
What is the operating point? H263 with low average quantizer size OR HVS-best with high average quantizer? It depends on the matrix. I only know SixOfNine quite well. IMHO SixOfNine is better than h.263 (or MPEG) up to the point where the quantizer of SixOfNine climbs up to 5. That means, use it if you aim for higher quality than what H.263 at quant=4 can deliver. But that's really at the fringes of what SixOfNine should be used for and in those ranges there are probably CQMs that are better up to the job.

The whole spectrum between highly and lowly compressing is of course covered by custom quantization matrices. Just try out all those well-tuned matrices by Sharktooth for example. And he's also done high- and very-high-bitrate matrices.

I only encode at 'high' to 'very high' bitrates, or more correctly 'low compression', so I'm sorry I cannot help you more than that. Have a look at the public CQM tests that Soulhunter performed.

[15081947]

It depends on the matrix. I only know SixOfNine quite well. IMHO SixOfNine is better than h.263 (or MPEG) up to the point where the quantizer of SixOfNine climbs up to 5. That means, use it if you aim for higher quality than what H.263 at quant=4 can deliver. But that's really at the fringes of what SixOfNine should be used for and in those ranges there are probably CQMs that are better up to the job.

Thank you. That is the pointer I was looking for.

[shon3i]

Have a look at the public CQM tests that Soulhunter performed Where is that. I cant find on the forum please link. btw Did you mean that is not necessarily to use any other matrix than h263 for xvid encodings and bitrate up to 1200kbs and less.

[Soulhunter]

Where is that. I cant find on the forum please link...
- CQM comparison V3 round 1 (http://forum.doom9.org/showthread.php?t=84375)

- CQM comparison V3 round 2 (http://forum.doom9.org/showthread.php?t=85455)

- CQM comparison V4 (http://forum.doom9.org/showthread.php?t=90785)


For more info do a search for "CQM*" <- dont forget the "*" ^^


Bye