Commit 891a7d4df6 for aom
commit 891a7d4df629a639d7db24b46da186e45cf5f003
Author: Satheesh Kumar <satheesh.kumar@ittiam.com>
Date: Mon Dec 15 16:38:40 2025 +0530
Tune sf disable_loop_restoration_luma for speed >= 4
In this patch, the speed feature 'disable_loop_restoration_luma' is
disabled and the speed features 'disable_sgr_filter' and
'disable_wiener_coeff_refine_search' are enabled. This change
improves the VMAF performance for speed >= 4.
Performance results of encoder and decoder for HDRES2 resolution.
Instruction
Count Reduction(%) BD-Rate Loss(%)
cpu Encoder Decoder avg.psnr ssim vmaf vmaf_neg
4 -0.115 -0.388 -0.0755 -0.0598 -0.4995 -0.3505
5 -0.222 -0.518 -0.0610 -0.0543 -0.3819 -0.2567
6 -0.324 -0.488 -0.0694 -0.0736 -0.3513 -0.2507
STATS_CHANGED for speed >= 4
Change-Id: I791113124afea90fc69b0d214da47f336b915b22
diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c
index 0a02a45a0f..05e1718107 100644
--- a/av1/encoder/speed_features.c
+++ b/av1/encoder/speed_features.c
@@ -2941,7 +2941,8 @@ void av1_set_speed_features_qindex_dependent(AV1_COMP *cpi, int speed) {
const int qindex_high[2] = { 180, 160 };
if (cm->quant_params.base_qindex <= qindex_low[is_720p_or_larger] ||
cm->quant_params.base_qindex > qindex_high[is_720p_or_larger]) {
- sf->lpf_sf.disable_loop_restoration_luma = 1;
+ sf->lpf_sf.disable_sgr_filter = true;
+ sf->lpf_sf.disable_wiener_coeff_refine_search = true;
}
}
}