vp9-svc: Use eightap_smooth for downsampling at low resol.

Switch from bilinear to eighttap_smooth for frame-level
downsampling at low resolutions (<= 320x240).

avgPSNR/SSIM metrics increase from ~0.5-2% (all clips positive gain),
for 2 and 3 spatial layer SVC, with 3 temporal layers.
Small/negligible increase in encoding time (< 1%).

Change-Id: I758472fc4fddd51d87f13c9d1a1cd4986ef5d41f

diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index 2636bd9a580205cecdbc58a087ec71cc56b222a3..4f137ecc396b9ddd7d406ea58b7f4e79e50de210 100644 (file)
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -661,6 +661,10 @@ int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
   // averaging filter), use bilinear for now.
   if (width * height <= 640 * 480) {
     cpi->svc.downsample_filter_type[cpi->svc.spatial_layer_id] = BILINEAR;
+    // Use Eightap_smooth for low resolutions.
+    if (width * height <= 320 * 240)
+      cpi->svc.downsample_filter_type[cpi->svc.spatial_layer_id] =
+          EIGHTTAP_SMOOTH;
     cpi->svc.downsample_filter_phase[cpi->svc.spatial_layer_id] = 8;
   }