summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--test/svc_datarate_test.cc76
-rw-r--r--vp9/encoder/vp9_ratectrl.c35
-rw-r--r--vp9/encoder/vp9_svc_layercontext.c34
-rw-r--r--vp9/encoder/vp9_svc_layercontext.h6
4 files changed, 134 insertions, 17 deletions
diff --git a/test/svc_datarate_test.cc b/test/svc_datarate_test.cc
index d6b247723..91cf9b73f 100644
--- a/test/svc_datarate_test.cc
+++ b/test/svc_datarate_test.cc
@@ -60,6 +60,7 @@ class DatarateOnePassCbrSvc : public OnePassCbrSvc {
memset(bits_total_, 0, sizeof(bits_total_));
memset(layer_target_avg_bandwidth_, 0, sizeof(layer_target_avg_bandwidth_));
dynamic_drop_layer_ = false;
+ single_layer_resize_ = false;
change_bitrate_ = false;
last_pts_ref_ = 0;
middle_bitrate_ = 0;
@@ -285,7 +286,7 @@ class DatarateOnePassCbrSvc : public OnePassCbrSvc {
encoder->Config(&cfg_);
}
- if (dynamic_drop_layer_) {
+ if (dynamic_drop_layer_ && !single_layer_resize_) {
// TODO(jian): Disable AQ Mode for this test for now.
encoder->Control(VP9E_SET_AQ_MODE, 0);
if (video->frame() == 0) {
@@ -329,8 +330,27 @@ class DatarateOnePassCbrSvc : public OnePassCbrSvc {
cfg_.rc_target_bitrate += cfg_.layer_target_bitrate[2];
encoder->Config(&cfg_);
}
+ } else if (dynamic_drop_layer_ && single_layer_resize_) {
+ // Change layer bitrates to set top layers to 0. This will trigger skip
+ // encoding/dropping of top spatial layers.
+ if (video->frame() == 10) {
+ cfg_.rc_target_bitrate -=
+ (cfg_.layer_target_bitrate[1] + cfg_.layer_target_bitrate[2]);
+ middle_bitrate_ = cfg_.layer_target_bitrate[1];
+ top_bitrate_ = cfg_.layer_target_bitrate[2];
+ cfg_.layer_target_bitrate[1] = 0;
+ cfg_.layer_target_bitrate[2] = 0;
+ // Set spatial layer 0 to a very low bitrate to trigger resize.
+ cfg_.layer_target_bitrate[0] = 30;
+ cfg_.rc_target_bitrate = cfg_.layer_target_bitrate[0];
+ encoder->Config(&cfg_);
+ } else if (video->frame() == 300) {
+ // Set base spatial layer to very high to go back up to original size.
+ cfg_.layer_target_bitrate[0] = 300;
+ cfg_.rc_target_bitrate = cfg_.layer_target_bitrate[0];
+ encoder->Config(&cfg_);
+ }
}
-
if (force_key_test_ && force_key_) frame_flags_ = VPX_EFLAG_FORCE_KF;
if (insert_layer_sync_) {
@@ -483,13 +503,15 @@ class DatarateOnePassCbrSvc : public OnePassCbrSvc {
}
}
- ASSERT_EQ(pkt->data.frame.width[sl],
- top_sl_width_ * svc_params_.scaling_factor_num[sl] /
- svc_params_.scaling_factor_den[sl]);
+ if (!single_layer_resize_) {
+ ASSERT_EQ(pkt->data.frame.width[sl],
+ top_sl_width_ * svc_params_.scaling_factor_num[sl] /
+ svc_params_.scaling_factor_den[sl]);
- ASSERT_EQ(pkt->data.frame.height[sl],
- top_sl_height_ * svc_params_.scaling_factor_num[sl] /
- svc_params_.scaling_factor_den[sl]);
+ ASSERT_EQ(pkt->data.frame.height[sl],
+ top_sl_height_ * svc_params_.scaling_factor_num[sl] /
+ svc_params_.scaling_factor_den[sl]);
+ }
}
}
@@ -525,6 +547,7 @@ class DatarateOnePassCbrSvc : public OnePassCbrSvc {
int tune_content_;
int spatial_layer_id_;
bool dynamic_drop_layer_;
+ bool single_layer_resize_;
unsigned int top_sl_width_;
unsigned int top_sl_height_;
vpx_svc_ref_frame_config_t ref_frame_config;
@@ -794,6 +817,43 @@ TEST_P(DatarateOnePassCbrSvcSingleBR, OnePassCbrSvc3SL_DisableEnableLayers) {
#endif
}
+// Check basic rate targeting for 1 pass CBR SVC with 2 spatial layers and on
+// the fly switching to 1 spatial layer with dynamic resize enabled.
+// The resizer will resize the single layer down and back up again, as the
+// bitrate goes back up.
+TEST_P(DatarateOnePassCbrSvcSingleBR, OnePassCbrSvc3SL_SingleLayerResize) {
+ SetSvcConfig(2, 1);
+ cfg_.rc_buf_initial_sz = 500;
+ cfg_.rc_buf_optimal_sz = 500;
+ cfg_.rc_buf_sz = 1000;
+ cfg_.rc_min_quantizer = 0;
+ cfg_.rc_max_quantizer = 63;
+ cfg_.g_threads = 1;
+ cfg_.temporal_layering_mode = 0;
+ cfg_.rc_dropframe_thresh = 30;
+ cfg_.kf_max_dist = 9999;
+ cfg_.rc_resize_allowed = 1;
+ ::libvpx_test::I420VideoSource video("niklas_640_480_30.yuv", 640, 480, 30, 1,
+ 0, 400);
+ top_sl_width_ = 640;
+ top_sl_height_ = 480;
+ cfg_.rc_target_bitrate = 800;
+ ResetModel();
+ dynamic_drop_layer_ = true;
+ single_layer_resize_ = true;
+ AssignLayerBitrates();
+ ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+ // Don't check rate targeting on two top spatial layer since they will be
+ // skipped for part of the sequence.
+ CheckLayerRateTargeting(number_spatial_layers_ - 2, number_temporal_layers_,
+ 0.78, 1.15);
+#if CONFIG_VP9_DECODER
+ // The non-reference frames are expected to be mismatched frames as the
+ // encoder will avoid loopfilter on these frames.
+ EXPECT_EQ(GetNonRefFrames(), GetMismatchFrames());
+#endif
+}
+
// Run SVC encoder for 1 temporal layer, 2 spatial layers, with spatial
// downscale 5x5.
TEST_P(DatarateOnePassCbrSvcSingleBR, OnePassCbrSvc2SL1TL5x5MultipleRuns) {
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index a5dbab631..edbed730b 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -2369,6 +2369,35 @@ void vp9_rc_get_svc_params(VP9_COMP *cpi) {
vp9_rc_set_frame_target(cpi, target);
if (cm->show_frame) update_buffer_level_svc_preencode(cpi);
+
+ if (cpi->oxcf.resize_mode == RESIZE_DYNAMIC && svc->single_layer_svc == 1 &&
+ svc->spatial_layer_id == svc->first_spatial_layer_to_encode) {
+ LAYER_CONTEXT *lc = NULL;
+ cpi->resize_pending = vp9_resize_one_pass_cbr(cpi);
+ if (cpi->resize_pending) {
+ int tl, width, height;
+ // Apply the same scale to all temporal layers.
+ for (tl = 0; tl < svc->number_temporal_layers; tl++) {
+ lc = &svc->layer_context[svc->spatial_layer_id *
+ svc->number_temporal_layers +
+ tl];
+ lc->scaling_factor_num_resize =
+ cpi->resize_scale_num * lc->scaling_factor_num;
+ lc->scaling_factor_den_resize =
+ cpi->resize_scale_den * lc->scaling_factor_den;
+ }
+ // Set the size for this current temporal layer.
+ lc = &svc->layer_context[svc->spatial_layer_id *
+ svc->number_temporal_layers +
+ svc->temporal_layer_id];
+ get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height,
+ lc->scaling_factor_num_resize,
+ lc->scaling_factor_den_resize, &width, &height);
+ vp9_set_size_literal(cpi, width, height);
+ }
+ } else {
+ cpi->resize_pending = 0;
+ }
}
void vp9_rc_get_one_pass_cbr_params(VP9_COMP *cpi) {
@@ -2625,7 +2654,7 @@ int vp9_resize_one_pass_cbr(VP9_COMP *cpi) {
int avg_qp_thr1 = 70;
int avg_qp_thr2 = 50;
int min_width = 180;
- int min_height = 180;
+ int min_height = 90;
int down_size_on = 1;
cpi->resize_scale_num = 1;
cpi->resize_scale_den = 1;
@@ -2635,6 +2664,7 @@ int vp9_resize_one_pass_cbr(VP9_COMP *cpi) {
cpi->resize_count = 0;
return 0;
}
+
// Check current frame reslution to avoid generating frames smaller than
// the minimum resolution.
if (ONEHALFONLY_RESIZE) {
@@ -2645,8 +2675,7 @@ int vp9_resize_one_pass_cbr(VP9_COMP *cpi) {
(cm->width * 3 / 4 < min_width || cm->height * 3 / 4 < min_height))
return 0;
else if (cpi->resize_state == THREE_QUARTER &&
- ((cpi->oxcf.width >> 1) < min_width ||
- (cpi->oxcf.height >> 1) < min_height))
+ (cm->width * 3 / 4 < min_width || cm->height * 3 / 4 < min_height))
down_size_on = 0;
}
diff --git a/vp9/encoder/vp9_svc_layercontext.c b/vp9/encoder/vp9_svc_layercontext.c
index 32ee6e064..1466d3a2b 100644
--- a/vp9/encoder/vp9_svc_layercontext.c
+++ b/vp9/encoder/vp9_svc_layercontext.c
@@ -55,6 +55,7 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
svc->use_set_ref_frame_config = 0;
svc->num_encoded_top_layer = 0;
svc->simulcast_mode = 0;
+ svc->single_layer_svc = 0;
for (i = 0; i < REF_FRAMES; ++i) {
svc->fb_idx_spatial_layer_id[i] = 0xff;
@@ -193,6 +194,7 @@ void vp9_update_layer_context_change_config(VP9_COMP *const cpi,
const RATE_CONTROL *const rc = &cpi->rc;
int sl, tl, layer = 0, spatial_layer_target;
float bitrate_alloc = 1.0;
+ int num_spatial_layers_nonzero_rate = 0;
cpi->svc.temporal_layering_mode = oxcf->temporal_layering_mode;
@@ -273,6 +275,17 @@ void vp9_update_layer_context_change_config(VP9_COMP *const cpi,
lrc->best_quality = rc->best_quality;
}
}
+ for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
+ // Check bitrate of spatia layer.
+ layer = LAYER_IDS_TO_IDX(sl, oxcf->ts_number_layers - 1,
+ oxcf->ts_number_layers);
+ if (oxcf->layer_target_bitrate[layer] > 0)
+ num_spatial_layers_nonzero_rate += 1;
+ }
+ if (num_spatial_layers_nonzero_rate == 1)
+ svc->single_layer_svc = 1;
+ else
+ svc->single_layer_svc = 0;
}
static LAYER_CONTEXT *get_layer_context(VP9_COMP *const cpi) {
@@ -751,6 +764,8 @@ int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
int width = 0, height = 0;
SVC *const svc = &cpi->svc;
LAYER_CONTEXT *lc = NULL;
+ int scaling_factor_num = 1;
+ int scaling_factor_den = 1;
svc->skip_enhancement_layer = 0;
if (svc->disable_inter_layer_pred == INTER_LAYER_PRED_OFF &&
@@ -888,18 +903,25 @@ int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
lrc->best_quality = vp9_quantizer_to_qindex(lc->min_q);
}
- get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height,
- lc->scaling_factor_num, lc->scaling_factor_den, &width,
- &height);
+ if (cpi->oxcf.resize_mode == RESIZE_DYNAMIC && svc->single_layer_svc == 1 &&
+ svc->spatial_layer_id == svc->first_spatial_layer_to_encode &&
+ cpi->resize_state != ORIG) {
+ scaling_factor_num = lc->scaling_factor_num_resize;
+ scaling_factor_den = lc->scaling_factor_den_resize;
+ } else {
+ scaling_factor_num = lc->scaling_factor_num;
+ scaling_factor_den = lc->scaling_factor_den;
+ }
+
+ get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height, scaling_factor_num,
+ scaling_factor_den, &width, &height);
// Use Eightap_smooth for low resolutions.
if (width * height <= 320 * 240)
svc->downsample_filter_type[svc->spatial_layer_id] = EIGHTTAP_SMOOTH;
// For scale factors > 0.75, set the phase to 0 (aligns decimated pixel
// to source pixel).
- lc = &svc->layer_context[svc->spatial_layer_id * svc->number_temporal_layers +
- svc->temporal_layer_id];
- if (lc->scaling_factor_num > (3 * lc->scaling_factor_den) >> 2)
+ if (scaling_factor_num > (3 * scaling_factor_den) >> 2)
svc->downsample_filter_phase[svc->spatial_layer_id] = 0;
// The usage of use_base_mv or partition_reuse assumes down-scale of 2x2.
diff --git a/vp9/encoder/vp9_svc_layercontext.h b/vp9/encoder/vp9_svc_layercontext.h
index f1ba77970..7e46500b5 100644
--- a/vp9/encoder/vp9_svc_layercontext.h
+++ b/vp9/encoder/vp9_svc_layercontext.h
@@ -47,6 +47,9 @@ typedef struct {
int min_q;
int scaling_factor_num;
int scaling_factor_den;
+ // Scaling factors used for internal resize scaling for single layer SVC.
+ int scaling_factor_num_resize;
+ int scaling_factor_den_resize;
TWO_PASS twopass;
vpx_fixed_buf_t rc_twopass_stats_in;
unsigned int current_video_frame_in_layer;
@@ -192,6 +195,9 @@ typedef struct SVC {
// Every spatial layer on a superframe whose base is key is key too.
int simulcast_mode;
+
+ // Flag to indicate SVC is dynamically switched to a single layer.
+ int single_layer_svc;
} SVC;
struct VP9_COMP;
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-10 21:22:09 +0000