[x264-devel] rate control for static macroblocks

[Anne Aaron]

> Probably because below a certain SATD, at any sane bitrate, the
> residual would be quantized to zero (resulting in a skip).

It's true that you don't want to corrupt the statistics with those of the skip macroblocks.
Then an improvement could be to change the complementary function that predicts the size based on the predictor. A special case could be added when the satd is low, so that we don't overestimate the bits for the low (non-zero) satd portions:

ratecontrol.c, line 1282
static double predict_size( predictor_t *p, double q, double var )
{
     if (var<10) return 0;
     return p->coeff*var / (q*p->count);
}