ref: a9b722e00c2df9a4a731c28521d43aeca5b3a035
parent: b758ba795a04d551caa3c1af2806cdf0c075803a
parent: 7dedf4cbc779f5b091513066745f54f46a3e71a5
author: Sai Deng <[email protected]>
date: Sat Apr 13 13:47:46 EDT 2019
Merge "Add Tune for SSIM"
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -259,6 +259,24 @@
x->mbmi_ext = x->mbmi_ext_base + (mi_row * cm->mi_cols + mi_col);
}
+static double get_ssim_rdmult_scaling_factor(VP9_COMP *const cpi, int mi_row,
+ int mi_col) {+ const VP9EncoderConfig *const oxcf = &cpi->oxcf;
+ if (oxcf->tuning == VP8_TUNE_SSIM) {+ const VP9_COMMON *const cm = &cpi->common;
+ // SSIM rdmult scaling factors are currently 64x64 based.
+ const int num_8x8_w = 8;
+ const int num_8x8_h = 8;
+ const int num_cols = (cm->mi_cols + num_8x8_w - 1) / num_8x8_w;
+ const int row = mi_row / num_8x8_h;
+ const int col = mi_col / num_8x8_w;
+ const int index = row * num_cols + col;
+
+ return cpi->mi_ssim_rdmult_scaling_factors[index];
+ }
+ return 1.0;
+}
+
static void set_offsets(VP9_COMP *cpi, const TileInfo *const tile,
MACROBLOCK *const x, int mi_row, int mi_col,
BLOCK_SIZE bsize) {@@ -267,6 +285,8 @@
const int mi_width = num_8x8_blocks_wide_lookup[bsize];
const int mi_height = num_8x8_blocks_high_lookup[bsize];
MvLimits *const mv_limits = &x->mv_limits;
+ const double ssim_factor =
+ get_ssim_rdmult_scaling_factor(cpi, mi_row, mi_col);
set_skip_context(xd, mi_row, mi_col);
@@ -293,6 +313,7 @@
// R/D setup.
x->rddiv = cpi->rd.RDDIV;
x->rdmult = cpi->rd.RDMULT;
+ x->rdmult = (int)(ssim_factor * x->rdmult);
// required by vp9_append_sub8x8_mvs_for_idx() and vp9_find_best_ref_mvs()
xd->tile = *tile;
@@ -1916,6 +1937,8 @@
VP9_COMMON *const cm = &cpi->common;
const uint8_t *const map =
cm->seg.update_map ? cpi->segmentation_map : cm->last_frame_seg_map;
+ const double ssim_factor =
+ get_ssim_rdmult_scaling_factor(cpi, mi_row, mi_col);
vp9_init_plane_quantizers(cpi, x);
vpx_clear_system_state();
@@ -1923,25 +1946,22 @@
if (aq_mode == NO_AQ || aq_mode == PSNR_AQ) {if (cpi->sf.enable_tpl_model || cpi->sf.enable_wiener_variance)
x->rdmult = x->cb_rdmult;
- return;
- }
-
- if (aq_mode == CYCLIC_REFRESH_AQ) {+ } else if (aq_mode == CYCLIC_REFRESH_AQ) {// If segment is boosted, use rdmult for that segment.
if (cyclic_refresh_segment_id_boosted(
get_segment_id(cm, map, bsize, mi_row, mi_col)))
x->rdmult = vp9_cyclic_refresh_get_rdmult(cpi->cyclic_refresh);
- return;
+ } else {+ x->rdmult = vp9_compute_rd_mult(cpi, cm->base_qindex + cm->y_dc_delta_q);
+ if (cpi->sf.enable_wiener_variance && cm->show_frame) {+ if (cm->seg.enabled)
+ x->rdmult = vp9_compute_rd_mult(
+ cpi, vp9_get_qindex(&cm->seg, x->e_mbd.mi[0]->segment_id,
+ cm->base_qindex));
+ }
}
- x->rdmult = vp9_compute_rd_mult(cpi, cm->base_qindex + cm->y_dc_delta_q);
-
- if (cpi->sf.enable_wiener_variance && cm->show_frame) {- if (cm->seg.enabled)
- x->rdmult = vp9_compute_rd_mult(
- cpi, vp9_get_qindex(&cm->seg, x->e_mbd.mi[0]->segment_id,
- cm->base_qindex));
- }
+ x->rdmult = (int)(ssim_factor * x->rdmult);
}
static void rd_pick_sb_modes(VP9_COMP *cpi, TileDataEnc *tile_data,
@@ -2179,8 +2199,12 @@
set_offsets(cpi, tile, x, mi_row, mi_col, bsize);
if ((cpi->sf.enable_tpl_model || cpi->sf.enable_wiener_variance) &&
- cpi->oxcf.aq_mode == NO_AQ)
+ cpi->oxcf.aq_mode == NO_AQ) {+ const double ssim_factor =
+ get_ssim_rdmult_scaling_factor(cpi, mi_row, mi_col);
x->rdmult = x->cb_rdmult;
+ x->rdmult = (int)(ssim_factor * x->rdmult);
+ }
update_state(cpi, td, ctx, mi_row, mi_col, bsize, output_enabled);
encode_superblock(cpi, td, tp, output_enabled, mi_row, mi_col, bsize, ctx);
@@ -3783,10 +3807,15 @@
int64_t dist_breakout_thr = cpi->sf.partition_search_breakout_thr.dist;
int rate_breakout_thr = cpi->sf.partition_search_breakout_thr.rate;
int must_split = 0;
- int partition_mul = x->cb_rdmult;
+
// Ref frames picked in the [i_th] quarter subblock during square partition
// RD search. It may be used to prune ref frame selection of rect partitions.
uint8_t ref_frames_used[4] = { 0, 0, 0, 0 };+
+ int partition_mul = x->cb_rdmult;
+ const double ssim_factor =
+ get_ssim_rdmult_scaling_factor(cpi, mi_row, mi_col);
+ partition_mul = (int)(ssim_factor * partition_mul);
(void)*tp_orig;
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -991,6 +991,9 @@
vpx_free(cpi->mb_wiener_variance);
cpi->mb_wiener_variance = NULL;
+ vpx_free(cpi->mi_ssim_rdmult_scaling_factors);
+ cpi->mi_ssim_rdmult_scaling_factors = NULL;
+
vp9_free_ref_frame_buffers(cm->buffer_pool);
#if CONFIG_VP9_POSTPROC
vp9_free_postproc_buffers(cm);
@@ -2389,6 +2392,17 @@
sizeof(*cpi->mb_wiener_variance)));
}
+ {+ const int bsize = BLOCK_64X64;
+ const int w = num_8x8_blocks_wide_lookup[bsize];
+ const int h = num_8x8_blocks_high_lookup[bsize];
+ const int num_cols = (cm->mi_cols + w - 1) / w;
+ const int num_rows = (cm->mi_rows + h - 1) / h;
+ CHECK_MEM_ERROR(cm, cpi->mi_ssim_rdmult_scaling_factors,
+ vpx_calloc(num_rows * num_cols,
+ sizeof(*cpi->mi_ssim_rdmult_scaling_factors)));
+ }
+
cpi->kmeans_data_arr_alloc = 0;
#if CONFIG_NON_GREEDY_MV
cpi->feature_score_loc_alloc = 0;
@@ -4718,6 +4732,72 @@
}
}
+// Implementation and modifications of C. Yeo, H. L. Tan, and Y. H. Tan, "On
+// rate distortion optimization using SSIM," Circuits and Systems for Video
+// Technology, IEEE Transactions on, vol. 23, no. 7, pp. 1170-1181, 2013.
+// SSIM_VAR_SCALE defines the strength of the bias towards SSIM in RDO.
+// Some sample values are:
+// SSIM_VAR_SCALE avg_psnr ssim ms_ssim (for midres test set)
+// 16.0 2.312 -3.062 -3.882
+// 32.0 0.852 -2.260 -2.821
+// 64.0 0.294 -1.606 -1.925
+#define SSIM_VAR_SCALE 16.0
+static void set_mb_ssim_rdmult_scaling(VP9_COMP *cpi) {+ const double c2 = 0.03 * 0.03 * 255 * 255;
+ VP9_COMMON *cm = &cpi->common;
+ ThreadData *td = &cpi->td;
+ MACROBLOCK *x = &td->mb;
+ MACROBLOCKD *xd = &x->e_mbd;
+ uint8_t *y_buffer = cpi->Source->y_buffer;
+ const int y_stride = cpi->Source->y_stride;
+ const int block_size = BLOCK_64X64;
+
+ const int num_8x8_w = num_8x8_blocks_wide_lookup[block_size];
+ const int num_8x8_h = num_8x8_blocks_high_lookup[block_size];
+ const int num_cols = (cm->mi_cols + num_8x8_w - 1) / num_8x8_w;
+ const int num_rows = (cm->mi_rows + num_8x8_h - 1) / num_8x8_h;
+ double log_sum = 0.0;
+ int row, col;
+
+ // Loop through each 64x64 block.
+ for (row = 0; row < num_rows; ++row) {+ for (col = 0; col < num_cols; ++col) {+ int mi_row, mi_col;
+ double var = 0.0, num_of_var = 0.0;
+ const int index = row * num_cols + col;
+
+ for (mi_row = row * num_8x8_h;
+ mi_row < cm->mi_rows && mi_row < (row + 1) * num_8x8_h; ++mi_row) {+ for (mi_col = col * num_8x8_w;
+ mi_col < cm->mi_cols && mi_col < (col + 1) * num_8x8_w; ++mi_col) {+ struct buf_2d buf;
+ const int row_offset_y = mi_row << 3;
+ const int col_offset_y = mi_col << 3;
+
+ buf.buf = y_buffer + row_offset_y * y_stride + col_offset_y;
+ buf.stride = y_stride;
+ var += vp9_get_sby_variance(cpi, &buf, BLOCK_8X8) / 64.0;
+ num_of_var += 1.0;
+ }
+ }
+ var = var / num_of_var / SSIM_VAR_SCALE;
+ var = 2.0 * var + c2;
+ cpi->mi_ssim_rdmult_scaling_factors[index] = var;
+ log_sum += log(var);
+ }
+ }
+ log_sum = exp(log_sum / (double)(num_rows * num_cols));
+
+ for (row = 0; row < num_rows; ++row) {+ for (col = 0; col < num_cols; ++col) {+ const int index = row * num_cols + col;
+ cpi->mi_ssim_rdmult_scaling_factors[index] /= log_sum;
+ }
+ }
+
+ (void)xd;
+}
+
// Process the wiener variance in 16x16 block basis.
static int qsort_comp(const void *elem1, const void *elem2) {int a = *((const int *)elem1);
@@ -4909,6 +4989,8 @@
cm->reset_frame_context = 2;
}
}
+
+ if (oxcf->tuning == VP8_TUNE_SSIM) set_mb_ssim_rdmult_scaling(cpi);
set_mb_wiener_variance(cpi);
--- a/vp9/encoder/vp9_encoder.h
+++ b/vp9/encoder/vp9_encoder.h
@@ -649,6 +649,7 @@
int64_t norm_wiener_variance;
int64_t *mb_wiener_variance;
+ double *mi_ssim_rdmult_scaling_factors;
int *stack_rank_buffer;
YV12_BUFFER_CONFIG last_frame_uf;
--- a/vp9/vp9_cx_iface.c
+++ b/vp9/vp9_cx_iface.c
@@ -263,9 +263,11 @@
RANGE_CHECK(extra_cfg, content, VP9E_CONTENT_DEFAULT,
VP9E_CONTENT_INVALID - 1);
- // TODO(yaowu): remove this when ssim tuning is implemented for vp9
+ // TODO(sdeng): remove this when ssim tuning is implemented for highbd
+#if CONFIG_VP9_HIGHBITDEPTH
if (extra_cfg->tuning == VP8_TUNE_SSIM)
- ERROR("Option --tune=ssim is not currently supported in VP9.");+ ERROR("Option --tune=ssim is not currently supported in highbd VP9.");+#endif
#if !CONFIG_REALTIME_ONLY
if (cfg->g_pass == VPX_RC_LAST_PASS) {