[XviD-devel] Re: qpel smearing/noise problem

[Christoph Lampert]

On Sun, 12 Jan 2003, suxen_drol wrote:
> using h.263 quant=1 with unnatural test sequences (ie. a grey image,
> with one sharp bright line) causes the articats to appear in qpel and
> hpel. i removed the quantization/dequantization stage from the
> encoder/decode, and the artifacts are gone.
> 
> this is the lossy-nature of mpeg-4, and not an xvid problem;
> coefficients are divided by two when using h.263 quant=1.

Did you remember the additional restriction on DCT/iDCT for quant=1 (which
we tried to get rid of by TOO_SMALL_LIMIT=2)? 
Something that a no blocks must be written to bytestream which 
may or may not lead to only-zero compensation after inverse quant and
iDCT? 

If I remember correctly, TOO_SMALL_LIMIT has been changed back to 1 in
some versions of XVID.

gruel