Add early termination in transform size search
In the full-rd transform size search, we go through all transform sizes to choose the one with best rd score. In this patch, an early termination is added to stop the search once we see that the smaller size won't give better rd score than the larger size. Also, the search starts from largest transform size, then goes down to smallest size. A speed feature tx_size_search_breakout is added, which is turned off at speed 0, and on for other speeds. The transform size search is turned on at speed 1. Borg test results: 1. At speed 1, derf set: psnr gain: 0.618%, ssim gain: 0.377%; stdhd set: psnr gain: 0.594%, ssim gain: 0.162%; No noticeable speed change. 3. At speed 2, derf set: psnr loss: 0.157%, ssim loss: 0.175%; stdhd set: psnr loss: 0.090%, ssim loss: 0.101%; speed gain: ~4%. Change-Id: I22535cd2017b5e54f2a62bb6a38231aea4268b3f
This commit is contained in:
Родитель
bac9beff72
Коммит
ba70f16011
|
@ -458,18 +458,18 @@ static void choose_tx_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
|
|||
{INT64_MAX, INT64_MAX},
|
||||
{INT64_MAX, INT64_MAX},
|
||||
{INT64_MAX, INT64_MAX}};
|
||||
TX_SIZE n, m;
|
||||
int n, m;
|
||||
int s0, s1;
|
||||
const TX_SIZE max_mode_tx_size = tx_mode_to_biggest_tx_size[cm->tx_mode];
|
||||
int64_t best_rd = INT64_MAX;
|
||||
TX_SIZE best_tx = TX_4X4;
|
||||
TX_SIZE best_tx = max_tx_size;
|
||||
|
||||
const vp9_prob *tx_probs = get_tx_probs2(max_tx_size, xd, &cm->fc.tx_probs);
|
||||
assert(skip_prob > 0);
|
||||
s0 = vp9_cost_bit(skip_prob, 0);
|
||||
s1 = vp9_cost_bit(skip_prob, 1);
|
||||
|
||||
for (n = TX_4X4; n <= max_tx_size; n++) {
|
||||
for (n = max_tx_size; n >= 0; n--) {
|
||||
txfm_rd_in_plane(x, &r[n][0], &d[n], &s[n],
|
||||
&sse[n], ref_best_rd, 0, bs, n,
|
||||
cpi->sf.use_fast_coef_costing);
|
||||
|
@ -491,6 +491,13 @@ static void choose_tx_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
|
|||
rd[n][1] = RDCOST(x->rdmult, x->rddiv, r[n][1] + s0, d[n]);
|
||||
}
|
||||
|
||||
// Early termination in transform size search.
|
||||
if (cpi->sf.tx_size_search_breakout &&
|
||||
(rd[n][1] == INT64_MAX ||
|
||||
(n < max_tx_size && rd[n][1] > rd[n + 1][1]) ||
|
||||
s[n] == 1))
|
||||
break;
|
||||
|
||||
if (rd[n][1] < best_rd) {
|
||||
best_tx = n;
|
||||
best_rd = rd[n][1];
|
||||
|
|
|
@ -59,8 +59,6 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
|
|||
if (speed >= 1) {
|
||||
sf->use_square_partition_only = !frame_is_intra_only(cm);
|
||||
sf->less_rectangular_check = 1;
|
||||
sf->tx_size_search_method = frame_is_boosted(cpi) ? USE_FULL_RD
|
||||
: USE_LARGESTALL;
|
||||
|
||||
if (MIN(cm->width, cm->height) >= 720)
|
||||
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
|
||||
|
@ -80,9 +78,14 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
|
|||
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
|
||||
sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V;
|
||||
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
|
||||
|
||||
sf->tx_size_search_breakout = 1;
|
||||
}
|
||||
|
||||
if (speed >= 2) {
|
||||
sf->tx_size_search_method = frame_is_boosted(cpi) ? USE_FULL_RD
|
||||
: USE_LARGESTALL;
|
||||
|
||||
if (MIN(cm->width, cm->height) >= 720) {
|
||||
sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
|
||||
sf->last_partitioning_redo_frequency = 3;
|
||||
|
@ -387,6 +390,7 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
|
|||
// Recode loop tolerence %.
|
||||
sf->recode_tolerance = 25;
|
||||
sf->default_interp_filter = SWITCHABLE;
|
||||
sf->tx_size_search_breakout = 0;
|
||||
|
||||
switch (oxcf->mode) {
|
||||
case ONE_PASS_BEST:
|
||||
|
|
|
@ -374,6 +374,10 @@ typedef struct SPEED_FEATURES {
|
|||
|
||||
// default interp filter choice
|
||||
INTERP_FILTER default_interp_filter;
|
||||
|
||||
// Early termination in transform size search, which only applies while
|
||||
// tx_size_search_method is USE_FULL_RD.
|
||||
int tx_size_search_breakout;
|
||||
} SPEED_FEATURES;
|
||||
|
||||
struct VP9_COMP;
|
||||
|
|
Загрузка…
Ссылка в новой задаче