Commit 1ae7ce7b12 for aom
commit 1ae7ce7b12d6758a05598278f3d9bc779fa17f4a
Author: Julio Barba <juliobbv@gmail.com>
Date: Wed Nov 26 13:59:03 2025 -0500
Adaptive sharpness: tweak LPF threshold
Testing QP 30 with tune=iq revealed a lower loop filter sharpness
strength was slightly more beneficial perceptually.
This change also comes with a 0.02 point SSIMULACRA 2 improvement
(at no file size change).
Bug: aomedia:463698815
Change-Id: I15f4429795db44dc8d65c24cd3662f65f813afb4
diff --git a/av1/encoder/picklpf.c b/av1/encoder/picklpf.c
index 613613d1e8..ecf4244b63 100644
--- a/av1/encoder/picklpf.c
+++ b/av1/encoder/picklpf.c
@@ -228,7 +228,7 @@ void av1_pick_filter_level(const YV12_BUFFER_CONFIG *sd, AV1_COMP *cpi,
// written to pick levels 0, 1 and 7 to keep it simple.
int max_lf_sharpness;
- if (cm->quant_params.base_qindex <= 120) {
+ if (cm->quant_params.base_qindex <= 112) {
max_lf_sharpness = 7;
} else if (cm->quant_params.base_qindex <= 160) {
max_lf_sharpness = 1;