H.263 optimized matrix

[rtjnyoface]

Hey does anyone have any idea what the actual matrix is? I'd like to try it out with xvid since xvid only has h.263 and MPEG. I can make my own matrices but was wondering if anyone knew the matrix to h.263 optimized. One last comment; I've signed up recently after about 2 years of constant browsing and after about 6 months at the official divx forum and I've noticed that quite a few people's questions go unanswered; some for even a few months/years. My only hope is that this forum is as friendly as the divx forum. Here's to a hopeful future here at this forum:).

[Manao]

If I'm not mistaken, h263 optimized is only h263 + trellis in the DivX nomenclature.

[Dark Shikari]

H.263 isn't a matrix per se; the "H.263" mode in MPEG-4 ASP doesn't allow custom quantization matrices and represents a mode with a built-in (both encoder and decoder side) large deadzone, resulting in significant smoothing-out of detail but lower bit cost.

"H.263 optimized" in DivX just sounds like a fancy name for some sort of quantization method.

Edit: Simulpost with Manao: yeah, it just sounds like its H.263+Trellis.

[DigitAl56K]

The H.263 Optimized mode is an enhancement on the implementation of the H.263 quantization routine that yields better PSNR but adds around 15% to your encoding time (if memory serves), so it's a nice option when you're encoding from a file source but you might want to avoid it if you're encoding a live stream. The output is fully compliant with H.263 inverse quantization, i.e. there is no alternative matrix used.

[rtjnyoface]

Thank you very much for your answer.It is amazing to see such advanced users such as yourself respond to simlple newby questions. I'm not an entire noob but as far as true technicalities of what happens and how it happens within a codec I have yet to find out. I don't quite understand how it can enhance the implementation of the h.263 quantization routine yet no alternative matrix is used. Would you be willing to explain?

[Dark Shikari]

Thank you very much for your answer.It is amazing to see such advanced users such as yourself respond to simlple newby questions. I'm not an entire noob but as far as true technicalities of what happens and how it happens within a codec I have yet to find out. I don't quite understand how it can enhance the implementation of the h.263 quantization routine yet no alternative matrix is used. Would you be willing to explain?When the DCT coefficients are quantized, you have a simple decision to make: for each one, do you round up or down?

The most obvious is to just round the way you were taught in grade school; round 0.6 to 1, 0.4 to 0, etc.

But you can do better--because of how the entropy coding in most modern video formats works, rounding down is better than rounding up, since it saves more bits. So, say, 0.7 and above go to 1, the rest go to 0 (and the same for higher numbers). This is called deadzone quant, and is used by most video encoders as a fast quant method.

But you can still do better--choose the rounding that maximizes quality relative to the bit cost of coding those coefficients. This is what trellis does.

[RickA]

I too had been wondering about this. Most helpful explanations by all indeed. Thanks for expressing this in layman's terms.

Cheers,
Rick