Further q estimate adjustment.
Slightly reduces the mean tendency to undershoot target rate in vbr, especially when using the memory less mode and when recodes are disabled. The effect is primarily at low q. Change-Id: I59a593b99522cc7da31b4134d1c8a65f5b7b7c53
This commit is contained in:
Родитель
d99e890840
Коммит
e1312c3696
|
@ -954,7 +954,7 @@ static double calc_correction_factor(double err_per_mb,
|
|||
const double error_term = err_per_mb / err_divisor;
|
||||
|
||||
// Adjustment based on actual quantizer to power term.
|
||||
const double power_term = MIN(vp9_convert_qindex_to_q(q) * 0.01 + pt_low,
|
||||
const double power_term = MIN(vp9_convert_qindex_to_q(q) * 0.0125 + pt_low,
|
||||
pt_high);
|
||||
|
||||
// Calculate correction factor
|
||||
|
@ -1014,7 +1014,7 @@ static int estimate_max_q(VP9_COMP *cpi,
|
|||
int bits_per_mb_at_this_q;
|
||||
|
||||
err_correction_factor = calc_correction_factor(err_per_mb,
|
||||
ERR_DIVISOR, 0.4, 0.90, q);
|
||||
ERR_DIVISOR, 0.5, 0.90, q);
|
||||
|
||||
bits_per_mb_at_this_q = vp9_rc_bits_per_mb(INTER_FRAME, q,
|
||||
err_correction_factor);
|
||||
|
@ -1064,7 +1064,7 @@ static int estimate_cq(VP9_COMP *cpi,
|
|||
|
||||
// Error per MB based correction factor
|
||||
err_correction_factor =
|
||||
calc_correction_factor(err_per_mb, 100.0, 0.4, 0.90, q) * clip_iifactor;
|
||||
calc_correction_factor(err_per_mb, 100.0, 0.5, 0.90, q) * clip_iifactor;
|
||||
|
||||
bits_per_mb_at_this_q =
|
||||
vp9_rc_bits_per_mb(INTER_FRAME, q, err_correction_factor);
|
||||
|
|
Загрузка…
Ссылка в новой задаче