From d36852b702e6726db57cabe19e47d9aaa18c3ecb Mon Sep 17 00:00:00 2001
From: Yunqing Wang <yunqingwang@google.com>
Date: Thu, 11 Jul 2013 11:15:00 -0700
Subject: [PATCH] Add encoding option --static-thresh
This option exists in VP8, and it was rewritten in VP9 to support
skipping on different partition levels. After prediction is done,
we can check if the residuals in the partition block will be all
quantized to 0. If this is true, the skip flag is set, and only
prediction data are needed in reconstruction. Based on DCT's energy
conservation property, the skipping check can be estimated in
spatial domain.
The prediction error is calculated and compared to a threshold.
The threshold is determined by the dequant values, and also
adjusted by partition sizes. To be precise, the DC and AC parts
for Y, U, and V planes are checked to decide skipping or not.
Test showed that
1. derf set:
when static-thresh = 1, psnr loss is 0.666%;
when static-thresh = 500, psnr loss is 1.162%;
2. stdhd set:
when static-thresh = 1, psnr loss is 1.249%;
when static-thresh = 500, psnr loss is 1.668%;
For different clips, encoding speedup range is between several
percentage and 20+% when static-thresh <= 500. For example,
clip bitrate static-thresh psnr time
akiyo(cif) 500 0 48.923 5.635s(50f)
akiyo 500 500 48.863 4.402s(50f)
parkjoy(1080p) 4000 0 30.380 77.54s(30f)
parkjoy 4000 500 30.384 69.59s(30f)
sunflower(1080p) 4000 0 44.461 85.2s(30f)
sunflower 4000 500 44.418 78.1s(30f)
Higher static-thresh values give larger speedup with larger
quality loss.
Change-Id: I857031ceb466ff314ab580ac5ec5d18542203c53
---
vp9/encoder/vp9_encodeframe.c | 3 +
vp9/encoder/vp9_rdopt.c | 131 +++++++++++++++++++++-------------
2 files changed, 84 insertions(+), 50 deletions(-)
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index f2cb478bea..ae1389b156 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -535,8 +535,11 @@ static void set_offsets(VP9_COMP *cpi, int mi_row, int mi_col,
cpi->seg0_progress = ((y * mb_cols + x * 4 + p32 + p16 + tile_progress)
<< 16) / cm->MBs;
}
+
+ x->encode_breakout = cpi->segment_encode_breakout[mbmi->segment_id];
} else {
mbmi->segment_id = 0;
+ x->encode_breakout = cpi->oxcf.encode_breakout;
}
}
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 4f5b90870c..5c61146689 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -2941,53 +2941,84 @@ static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
if (cpi->common.mcomp_filter_type == SWITCHABLE)
*rate2 += get_switchable_rate(cm, x);
- if (cpi->active_map_enabled && x->active_ptr[0] == 0)
- x->skip = 1;
- else if (x->encode_breakout) {
- const BLOCK_SIZE_TYPE y_size = get_plane_block_size(bsize, &xd->plane[0]);
- const BLOCK_SIZE_TYPE uv_size = get_plane_block_size(bsize, &xd->plane[1]);
-
- unsigned int var, sse;
- int threshold = (xd->plane[0].dequant[1] * xd->plane[0].dequant[1] >> 4);
-
-
- if (threshold < x->encode_breakout)
- threshold = x->encode_breakout;
-
- var = cpi->fn_ptr[y_size].vf(x->plane[0].src.buf, x->plane[0].src.stride,
- xd->plane[0].dst.buf, xd->plane[0].dst.stride,
- &sse);
-
- if ((int)sse < threshold) {
- unsigned int q2dc = xd->plane[0].dequant[0];
- // If there is no codeable 2nd order dc
- // or a very small uniform pixel change change
- if ((sse - var < q2dc * q2dc >> 4) ||
- (sse / 2 > var && sse - var < 64)) {
- // Check u and v to make sure skip is ok
- int sse2;
- unsigned int sse2u, sse2v;
- var = cpi->fn_ptr[uv_size].vf(x->plane[1].src.buf,
- x->plane[1].src.stride,
- xd->plane[1].dst.buf,
- xd->plane[1].dst.stride, &sse2u);
- var = cpi->fn_ptr[uv_size].vf(x->plane[2].src.buf,
- x->plane[2].src.stride,
- xd->plane[2].dst.buf,
- xd->plane[2].dst.stride, &sse2v);
- sse2 = sse2u + sse2v;
-
- if (sse2 * 2 < threshold) {
- x->skip = 1;
- *distortion = sse + sse2;
- *rate2 = 500;
-
- // for best yrd calculation
- *rate_uv = 0;
- *distortion_uv = sse2;
-
- *disable_skip = 1;
- this_rd = RDCOST(x->rdmult, x->rddiv, *rate2, *distortion);
+ if (!is_comp_pred) {
+ if (cpi->active_map_enabled && x->active_ptr[0] == 0)
+ x->skip = 1;
+ else if (x->encode_breakout) {
+ const BLOCK_SIZE_TYPE y_size = get_plane_block_size(bsize, &xd->plane[0]);
+ const BLOCK_SIZE_TYPE uv_size = get_plane_block_size(bsize,
+ &xd->plane[1]);
+ unsigned int var, sse;
+ // Skipping threshold for ac.
+ unsigned int thresh_ac;
+ // The encode_breakout input
+ unsigned int encode_breakout = x->encode_breakout << 4;
+
+ // Adjust threshold according to dequant value. Making threshold more
+ // strict when dequant is high to avoid big PSNR loss.
+ if (xd->plane[0].dequant[1] < 32)
+ // Initial threshold value
+ thresh_ac = xd->plane[0].dequant[1] * xd->plane[0].dequant[1];
+ else
+ thresh_ac = (xd->plane[0].dequant[1] << 6) - 1024;
+
+ if (thresh_ac > 36000)
+ thresh_ac = 36000;
+
+ // Use encode_breakout input if it is bigger than internal threshold.
+ if (thresh_ac < encode_breakout)
+ thresh_ac = encode_breakout;
+
+ var = cpi->fn_ptr[y_size].vf(x->plane[0].src.buf, x->plane[0].src.stride,
+ xd->plane[0].dst.buf,
+ xd->plane[0].dst.stride, &sse);
+
+ // Adjust threshold according to partition size.
+ thresh_ac >>= 8 - (b_width_log2_lookup[bsize] +
+ b_height_log2_lookup[bsize]);
+
+ // Y skipping condition checking
+ if (sse < thresh_ac || sse == 0) {
+ // Skipping threshold for dc
+ unsigned int thresh_dc;
+
+ thresh_dc = (xd->plane[0].dequant[0] * xd->plane[0].dequant[0] >> 6);
+
+ // dc skipping checking
+ if ((sse - var) < thresh_dc || sse == var) {
+ unsigned int sse_u, sse_v;
+ unsigned int var_u, var_v;
+
+ var_u = cpi->fn_ptr[uv_size].vf(x->plane[1].src.buf,
+ x->plane[1].src.stride,
+ xd->plane[1].dst.buf,
+ xd->plane[1].dst.stride, &sse_u);
+
+ // U skipping condition checking
+ if ((sse_u * 4 < thresh_ac || sse_u == 0) &&
+ (sse_u - var_u < thresh_dc || sse_u == var_u)) {
+ var_v = cpi->fn_ptr[uv_size].vf(x->plane[2].src.buf,
+ x->plane[2].src.stride,
+ xd->plane[2].dst.buf,
+ xd->plane[2].dst.stride, &sse_v);
+
+ // V skipping condition checking
+ if ((sse_v * 4 < thresh_ac || sse_v == 0) &&
+ (sse_v - var_v < thresh_dc || sse_v == var_v)) {
+ x->skip = 1;
+
+ *rate2 = 500;
+ *rate_uv = 0;
+
+ // Scaling factor for SSE from spatial domain to frequency domain
+ // is 16. Adjust distortion accordingly.
+ *distortion_uv = (sse_u + sse_v) << 4;
+ *distortion = (sse << 4) + *distortion_uv;
+
+ *disable_skip = 1;
+ this_rd = RDCOST(x->rdmult, x->rddiv, *rate2, *distortion);
+ }
+ }
}
}
}
@@ -3236,6 +3267,7 @@ int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
for (i = 0; i < NB_TXFM_MODES; ++i)
txfm_cache[i] = INT64_MAX;
+ x->skip = 0;
this_mode = vp9_mode_order[mode_index].mode;
ref_frame = vp9_mode_order[mode_index].ref_frame;
@@ -3261,8 +3293,6 @@ int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
vp9_segfeature_active(&xd->seg, segment_id, SEG_LVL_REF_FRAME))
continue;
- x->skip = 0;
-
// Skip some checking based on small partitions' result.
if (x->fast_ms > 1 && !ref_frame)
continue;
@@ -3885,9 +3915,10 @@ int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
if (early_term)
break;
- if (x->skip && !mode_excluded)
+ if (x->skip && !comp_pred)
break;
}
+
if (best_rd >= best_rd_so_far)
return INT64_MAX;
--
GitLab