Support layer skipping in key svc structures

Add KSvc structure for 3 spatial and 3 temporal layers to allow to cover
more scenarious

Bug: webrtc:11999
Change-Id: Id16d1acfb4ca5f98d1b17d8f66d54b31d22d0745
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/188122
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Philip Eliasson <philipel@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32434}
This commit is contained in:
Danil Chapovalov
2020-10-16 16:14:58 +02:00
committed by Commit Bot
parent 609b047b07
commit 4b18e24967
12 changed files with 641 additions and 372 deletions

View File

@ -32,20 +32,18 @@ rtc_source_set("scalability_structures") {
"create_scalability_structure.h",
"scalability_structure_full_svc.cc",
"scalability_structure_full_svc.h",
"scalability_structure_key_svc.cc",
"scalability_structure_key_svc.h",
"scalability_structure_l1t2.cc",
"scalability_structure_l1t2.h",
"scalability_structure_l1t3.cc",
"scalability_structure_l1t3.h",
"scalability_structure_l2t1.cc",
"scalability_structure_l2t1.h",
"scalability_structure_l2t1_key.cc",
"scalability_structure_l2t1_key.h",
"scalability_structure_l2t1h.cc",
"scalability_structure_l2t1h.h",
"scalability_structure_l2t2.cc",
"scalability_structure_l2t2.h",
"scalability_structure_l2t2_key.cc",
"scalability_structure_l2t2_key.h",
"scalability_structure_l2t2_key_shift.cc",
"scalability_structure_l2t2_key_shift.h",
"scalability_structure_l3t1.cc",
@ -58,6 +56,7 @@ rtc_source_set("scalability_structures") {
deps = [
":scalable_video_controller",
"../../../api/transport/rtp:dependency_descriptor",
"../../../api/video:video_bitrate_allocation",
"../../../common_video/generic_frame_descriptor",
"../../../rtc_base:checks",
"../../../rtc_base:logging",
@ -73,6 +72,7 @@ if (rtc_include_tests) {
rtc_library("scalability_structure_tests") {
testonly = true
sources = [
"scalability_structure_key_svc_unittest.cc",
"scalability_structure_l3t3_unittest.cc",
"scalability_structure_test_helpers.cc",
"scalability_structure_test_helpers.h",
@ -83,6 +83,7 @@ if (rtc_include_tests) {
":scalable_video_controller",
"..:chain_diff_calculator",
"..:frame_dependencies_calculator",
"../../../api:array_view",
"../../../api/transport/rtp:dependency_descriptor",
"../../../api/video:video_bitrate_allocation",
"../../../api/video:video_frame_type",

View File

@ -12,13 +12,12 @@
#include <memory>
#include "absl/strings/string_view.h"
#include "modules/video_coding/svc/scalability_structure_key_svc.h"
#include "modules/video_coding/svc/scalability_structure_l1t2.h"
#include "modules/video_coding/svc/scalability_structure_l1t3.h"
#include "modules/video_coding/svc/scalability_structure_l2t1.h"
#include "modules/video_coding/svc/scalability_structure_l2t1_key.h"
#include "modules/video_coding/svc/scalability_structure_l2t1h.h"
#include "modules/video_coding/svc/scalability_structure_l2t2.h"
#include "modules/video_coding/svc/scalability_structure_l2t2_key.h"
#include "modules/video_coding/svc/scalability_structure_l2t2_key_shift.h"
#include "modules/video_coding/svc/scalability_structure_l3t1.h"
#include "modules/video_coding/svc/scalability_structure_l3t3.h"
@ -54,6 +53,7 @@ constexpr NamedStructureFactory kFactories[] = {
{"L2T2_KEY_SHIFT", Create<ScalabilityStructureL2T2KeyShift>},
{"L3T1", Create<ScalabilityStructureL3T1>},
{"L3T3", Create<ScalabilityStructureL3T3>},
{"L3T3_KEY", Create<ScalabilityStructureL3T3Key>},
{"S2T1", Create<ScalabilityStructureS2T1>},
};

View File

@ -0,0 +1,336 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "modules/video_coding/svc/scalability_structure_key_svc.h"
#include <bitset>
#include <utility>
#include <vector>
#include "absl/types/optional.h"
#include "api/transport/rtp/dependency_descriptor.h"
#include "api/video/video_bitrate_allocation.h"
#include "common_video/generic_frame_descriptor/generic_frame_info.h"
#include "modules/video_coding/svc/scalable_video_controller.h"
#include "rtc_base/checks.h"
#include "rtc_base/logging.h"
namespace webrtc {
namespace {
// Values to use as LayerFrameConfig::Id
enum : int { kKey, kDelta };
DecodeTargetIndication
Dti(int sid, int tid, const ScalableVideoController::LayerFrameConfig& config) {
if (config.IsKeyframe() || config.Id() == kKey) {
RTC_DCHECK_EQ(config.TemporalId(), 0);
return sid < config.SpatialId() ? DecodeTargetIndication::kNotPresent
: DecodeTargetIndication::kSwitch;
}
if (sid != config.SpatialId() || tid < config.TemporalId()) {
return DecodeTargetIndication::kNotPresent;
}
if (tid == config.TemporalId() && tid > 0) {
return DecodeTargetIndication::kDiscardable;
}
return DecodeTargetIndication::kSwitch;
}
} // namespace
constexpr int ScalabilityStructureKeySvc::kMaxNumSpatialLayers;
constexpr int ScalabilityStructureKeySvc::kMaxNumTemporalLayers;
ScalabilityStructureKeySvc::ScalabilityStructureKeySvc(int num_spatial_layers,
int num_temporal_layers)
: num_spatial_layers_(num_spatial_layers),
num_temporal_layers_(num_temporal_layers),
active_decode_targets_(
(uint32_t{1} << (num_spatial_layers * num_temporal_layers)) - 1) {
// There is no point to use this structure without spatial scalability.
RTC_DCHECK_GT(num_spatial_layers, 1);
RTC_DCHECK_LE(num_spatial_layers, kMaxNumSpatialLayers);
RTC_DCHECK_LE(num_temporal_layers, kMaxNumTemporalLayers);
}
ScalabilityStructureKeySvc::~ScalabilityStructureKeySvc() = default;
ScalableVideoController::StreamLayersConfig
ScalabilityStructureKeySvc::StreamConfig() const {
StreamLayersConfig result;
result.num_spatial_layers = num_spatial_layers_;
result.num_temporal_layers = num_temporal_layers_;
result.scaling_factor_num[num_spatial_layers_ - 1] = 1;
result.scaling_factor_den[num_spatial_layers_ - 1] = 1;
for (int sid = num_spatial_layers_ - 1; sid > 0; --sid) {
result.scaling_factor_num[sid - 1] = 1;
result.scaling_factor_den[sid - 1] = 2 * result.scaling_factor_den[sid];
}
return result;
}
bool ScalabilityStructureKeySvc::TemporalLayerIsActive(int tid) const {
if (tid >= num_temporal_layers_) {
return false;
}
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
if (DecodeTargetIsActive(sid, tid)) {
return true;
}
}
return false;
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureKeySvc::KeyframeConfig() {
std::vector<LayerFrameConfig> configs;
configs.reserve(num_spatial_layers_);
absl::optional<int> spatial_dependency_buffer_id;
spatial_id_is_enabled_.reset();
// Disallow temporal references cross T0 on higher temporal layers.
can_reference_t1_frame_for_spatial_id_.reset();
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
if (!DecodeTargetIsActive(sid, /*tid=*/0)) {
continue;
}
configs.emplace_back();
ScalableVideoController::LayerFrameConfig& config = configs.back();
config.Id(kKey).S(sid).T(0);
if (spatial_dependency_buffer_id) {
config.Reference(*spatial_dependency_buffer_id);
} else {
config.Keyframe();
}
config.Update(BufferIndex(sid, /*tid=*/0));
spatial_id_is_enabled_.set(sid);
spatial_dependency_buffer_id = BufferIndex(sid, /*tid=*/0);
}
return configs;
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureKeySvc::T0Config() {
std::vector<LayerFrameConfig> configs;
configs.reserve(num_spatial_layers_);
// Disallow temporal references cross T0 on higher temporal layers.
can_reference_t1_frame_for_spatial_id_.reset();
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
if (!DecodeTargetIsActive(sid, /*tid=*/0)) {
spatial_id_is_enabled_.reset(sid);
continue;
}
configs.emplace_back();
configs.back().Id(kDelta).S(sid).T(0).ReferenceAndUpdate(
BufferIndex(sid, /*tid=*/0));
}
return configs;
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureKeySvc::T1Config() {
std::vector<LayerFrameConfig> configs;
configs.reserve(num_spatial_layers_);
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
if (!DecodeTargetIsActive(sid, /*tid=*/1)) {
continue;
}
configs.emplace_back();
ScalableVideoController::LayerFrameConfig& config = configs.back();
config.Id(kDelta).S(sid).T(1).Reference(BufferIndex(sid, /*tid=*/0));
if (num_temporal_layers_ > 2) {
config.Update(BufferIndex(sid, /*tid=*/1));
can_reference_t1_frame_for_spatial_id_.set(sid);
}
}
return configs;
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureKeySvc::T2Config() {
std::vector<LayerFrameConfig> configs;
configs.reserve(num_spatial_layers_);
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
if (!DecodeTargetIsActive(sid, /*tid=*/2)) {
continue;
}
configs.emplace_back();
ScalableVideoController::LayerFrameConfig& config = configs.back();
config.Id(kDelta).S(sid).T(2);
if (can_reference_t1_frame_for_spatial_id_[sid]) {
config.Reference(BufferIndex(sid, /*tid=*/1));
} else {
config.Reference(BufferIndex(sid, /*tid=*/0));
}
}
return configs;
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureKeySvc::NextFrameConfig(bool restart) {
if (active_decode_targets_.none()) {
last_pattern_ = kNone;
return {};
}
if (restart) {
last_pattern_ = kNone;
}
switch (last_pattern_) {
case kNone:
last_pattern_ = kDeltaT0;
return KeyframeConfig();
case kDeltaT2B:
last_pattern_ = kDeltaT0;
return T0Config();
case kDeltaT2A:
if (TemporalLayerIsActive(1)) {
last_pattern_ = kDeltaT1;
return T1Config();
}
last_pattern_ = kDeltaT0;
return T0Config();
case kDeltaT1:
if (TemporalLayerIsActive(2)) {
last_pattern_ = kDeltaT2B;
return T2Config();
}
last_pattern_ = kDeltaT0;
return T0Config();
case kDeltaT0:
if (TemporalLayerIsActive(2)) {
last_pattern_ = kDeltaT2A;
return T2Config();
} else if (TemporalLayerIsActive(1)) {
last_pattern_ = kDeltaT1;
return T1Config();
}
last_pattern_ = kDeltaT0;
return T0Config();
}
RTC_NOTREACHED();
return {};
}
GenericFrameInfo ScalabilityStructureKeySvc::OnEncodeDone(
const LayerFrameConfig& config) {
GenericFrameInfo frame_info;
frame_info.spatial_id = config.SpatialId();
frame_info.temporal_id = config.TemporalId();
frame_info.encoder_buffers = config.Buffers();
frame_info.decode_target_indications.reserve(num_spatial_layers_ *
num_temporal_layers_);
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
for (int tid = 0; tid < num_temporal_layers_; ++tid) {
frame_info.decode_target_indications.push_back(Dti(sid, tid, config));
}
}
frame_info.part_of_chain.assign(num_spatial_layers_, false);
if (config.IsKeyframe() || config.Id() == kKey) {
RTC_DCHECK_EQ(config.TemporalId(), 0);
for (int sid = config.SpatialId(); sid < num_spatial_layers_; ++sid) {
frame_info.part_of_chain[sid] = true;
}
} else if (config.TemporalId() == 0) {
frame_info.part_of_chain[config.SpatialId()] = true;
}
frame_info.active_decode_targets = active_decode_targets_;
return frame_info;
}
void ScalabilityStructureKeySvc::OnRatesUpdated(
const VideoBitrateAllocation& bitrates) {
for (int sid = 0; sid < num_spatial_layers_; ++sid) {
// Enable/disable spatial layers independetely.
bool active = bitrates.GetBitrate(sid, /*tid=*/0) > 0;
SetDecodeTargetIsActive(sid, /*tid=*/0, active);
if (!spatial_id_is_enabled_[sid] && active) {
// Key frame is required to reenable any spatial layer.
last_pattern_ = kNone;
}
for (int tid = 1; tid < num_temporal_layers_; ++tid) {
// To enable temporal layer, require bitrates for lower temporal layers.
active = active && bitrates.GetBitrate(sid, tid) > 0;
SetDecodeTargetIsActive(sid, tid, active);
}
}
}
ScalabilityStructureL2T1Key::~ScalabilityStructureL2T1Key() = default;
FrameDependencyStructure ScalabilityStructureL2T1Key::DependencyStructure()
const {
FrameDependencyStructure structure;
structure.num_decode_targets = 2;
structure.num_chains = 2;
structure.decode_target_protected_by_chain = {0, 1};
structure.templates.resize(4);
structure.templates[0].S(0).Dtis("S-").ChainDiffs({2, 1}).FrameDiffs({2});
structure.templates[1].S(0).Dtis("SS").ChainDiffs({0, 0});
structure.templates[2].S(1).Dtis("-S").ChainDiffs({1, 2}).FrameDiffs({2});
structure.templates[3].S(1).Dtis("-S").ChainDiffs({1, 1}).FrameDiffs({1});
return structure;
}
ScalabilityStructureL2T2Key::~ScalabilityStructureL2T2Key() = default;
FrameDependencyStructure ScalabilityStructureL2T2Key::DependencyStructure()
const {
FrameDependencyStructure structure;
structure.num_decode_targets = 4;
structure.num_chains = 2;
structure.decode_target_protected_by_chain = {0, 0, 1, 1};
structure.templates.resize(6);
auto& templates = structure.templates;
templates[0].S(0).T(0).Dtis("SSSS").ChainDiffs({0, 0});
templates[1].S(0).T(0).Dtis("SS--").ChainDiffs({4, 3}).FrameDiffs({4});
templates[2].S(0).T(1).Dtis("-D--").ChainDiffs({2, 1}).FrameDiffs({2});
templates[3].S(1).T(0).Dtis("--SS").ChainDiffs({1, 1}).FrameDiffs({1});
templates[4].S(1).T(0).Dtis("--SS").ChainDiffs({1, 4}).FrameDiffs({4});
templates[5].S(1).T(1).Dtis("---D").ChainDiffs({3, 2}).FrameDiffs({2});
return structure;
}
ScalabilityStructureL3T3Key::~ScalabilityStructureL3T3Key() = default;
FrameDependencyStructure ScalabilityStructureL3T3Key::DependencyStructure()
const {
FrameDependencyStructure structure;
structure.num_decode_targets = 9;
structure.num_chains = 3;
structure.decode_target_protected_by_chain = {0, 0, 0, 1, 1, 1, 2, 2, 2};
auto& t = structure.templates;
t.resize(15);
// Templates are shown in the order frames following them appear in the
// stream, but in `structure.templates` array templates are sorted by
// (`spatial_id`, `temporal_id`) since that is a dependency descriptor
// requirement. Indexes are written in hex for nicer alignment.
t[0x0].S(0).T(0).Dtis("SSSSSSSSS").ChainDiffs({0, 0, 0});
t[0x5].S(1).T(0).Dtis("---SSSSSS").ChainDiffs({1, 1, 1}).FrameDiffs({1});
t[0xA].S(2).T(0).Dtis("------SSS").ChainDiffs({2, 1, 1}).FrameDiffs({1});
t[0x3].S(0).T(2).Dtis("--D------").ChainDiffs({3, 2, 1}).FrameDiffs({3});
t[0x8].S(1).T(2).Dtis("-----D---").ChainDiffs({4, 3, 2}).FrameDiffs({3});
t[0xD].S(2).T(2).Dtis("--------D").ChainDiffs({5, 4, 3}).FrameDiffs({3});
t[0x2].S(0).T(1).Dtis("-DS------").ChainDiffs({6, 5, 4}).FrameDiffs({6});
t[0x7].S(1).T(1).Dtis("----DS---").ChainDiffs({7, 6, 5}).FrameDiffs({6});
t[0xC].S(2).T(1).Dtis("-------DS").ChainDiffs({8, 7, 6}).FrameDiffs({6});
t[0x4].S(0).T(2).Dtis("--D------").ChainDiffs({9, 8, 7}).FrameDiffs({3});
t[0x9].S(1).T(2).Dtis("-----D---").ChainDiffs({10, 9, 8}).FrameDiffs({3});
t[0xE].S(2).T(2).Dtis("--------D").ChainDiffs({11, 10, 9}).FrameDiffs({3});
t[0x1].S(0).T(0).Dtis("SSS------").ChainDiffs({12, 11, 10}).FrameDiffs({12});
t[0x6].S(1).T(0).Dtis("---SSS---").ChainDiffs({1, 12, 11}).FrameDiffs({12});
t[0xB].S(2).T(0).Dtis("------SSS").ChainDiffs({2, 1, 12}).FrameDiffs({12});
return structure;
}
} // namespace webrtc

View File

@ -0,0 +1,107 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_KEY_SVC_H_
#define MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_KEY_SVC_H_
#include <bitset>
#include <vector>
#include "api/transport/rtp/dependency_descriptor.h"
#include "api/video/video_bitrate_allocation.h"
#include "common_video/generic_frame_descriptor/generic_frame_info.h"
#include "modules/video_coding/svc/scalable_video_controller.h"
namespace webrtc {
class ScalabilityStructureKeySvc : public ScalableVideoController {
public:
ScalabilityStructureKeySvc(int num_spatial_layers, int num_temporal_layers);
~ScalabilityStructureKeySvc() override;
StreamLayersConfig StreamConfig() const override;
std::vector<LayerFrameConfig> NextFrameConfig(bool restart) override;
GenericFrameInfo OnEncodeDone(const LayerFrameConfig& config) override;
void OnRatesUpdated(const VideoBitrateAllocation& bitrates) override;
private:
enum FramePattern {
kNone,
kDeltaT0,
kDeltaT2A,
kDeltaT1,
kDeltaT2B,
};
static constexpr int kMaxNumSpatialLayers = 3;
static constexpr int kMaxNumTemporalLayers = 3;
// Index of the buffer to store last frame for layer (`sid`, `tid`)
int BufferIndex(int sid, int tid) const {
return tid * num_spatial_layers_ + sid;
}
bool DecodeTargetIsActive(int sid, int tid) const {
return active_decode_targets_[sid * num_temporal_layers_ + tid];
}
void SetDecodeTargetIsActive(int sid, int tid, bool value) {
active_decode_targets_.set(sid * num_temporal_layers_ + tid, value);
}
bool TemporalLayerIsActive(int tid) const;
std::vector<LayerFrameConfig> KeyframeConfig();
std::vector<LayerFrameConfig> T0Config();
std::vector<LayerFrameConfig> T1Config();
std::vector<LayerFrameConfig> T2Config();
const int num_spatial_layers_;
const int num_temporal_layers_;
FramePattern last_pattern_ = kNone;
std::bitset<kMaxNumSpatialLayers> spatial_id_is_enabled_;
std::bitset<kMaxNumSpatialLayers> can_reference_t1_frame_for_spatial_id_;
std::bitset<32> active_decode_targets_;
};
// S1 0--0--0-
// | ...
// S0 0--0--0-
class ScalabilityStructureL2T1Key : public ScalabilityStructureKeySvc {
public:
ScalabilityStructureL2T1Key() : ScalabilityStructureKeySvc(2, 1) {}
~ScalabilityStructureL2T1Key() override;
FrameDependencyStructure DependencyStructure() const override;
};
// S1T1 0 0
// / / /
// S1T0 0---0---0
// | ...
// S0T1 | 0 0
// |/ / /
// S0T0 0---0---0
// Time-> 0 1 2 3 4
class ScalabilityStructureL2T2Key : public ScalabilityStructureKeySvc {
public:
ScalabilityStructureL2T2Key() : ScalabilityStructureKeySvc(2, 2) {}
~ScalabilityStructureL2T2Key() override;
FrameDependencyStructure DependencyStructure() const override;
};
class ScalabilityStructureL3T3Key : public ScalabilityStructureKeySvc {
public:
ScalabilityStructureL3T3Key() : ScalabilityStructureKeySvc(3, 3) {}
~ScalabilityStructureL3T3Key() override;
FrameDependencyStructure DependencyStructure() const override;
};
} // namespace webrtc
#endif // MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_KEY_SVC_H_

View File

@ -0,0 +1,128 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "modules/video_coding/svc/scalability_structure_key_svc.h"
#include <vector>
#include "api/array_view.h"
#include "api/transport/rtp/dependency_descriptor.h"
#include "common_video/generic_frame_descriptor/generic_frame_info.h"
#include "modules/video_coding/svc/scalability_structure_test_helpers.h"
#include "test/gmock.h"
#include "test/gtest.h"
namespace webrtc {
namespace {
using ::testing::ElementsAre;
using ::testing::IsEmpty;
using ::testing::SizeIs;
TEST(ScalabilityStructureL3T3KeyTest,
SkipingT1FrameOnOneSpatialLayerKeepsStructureValid) {
ScalabilityStructureL3T3Key structure;
ScalabilityStructureWrapper wrapper(structure);
std::vector<GenericFrameInfo> frames;
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/3, /*s1=*/3));
wrapper.GenerateFrames(/*num_temporal_units=*/2, frames);
EXPECT_THAT(frames, SizeIs(4));
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/3, /*s1=*/1));
wrapper.GenerateFrames(/*num_temporal_units=*/1, frames);
EXPECT_THAT(frames, SizeIs(5));
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/3, /*s1=*/3));
wrapper.GenerateFrames(/*num_temporal_units=*/1, frames);
ASSERT_THAT(frames, SizeIs(7));
EXPECT_EQ(frames[0].temporal_id, 0);
EXPECT_EQ(frames[1].temporal_id, 0);
EXPECT_EQ(frames[2].temporal_id, 2);
EXPECT_EQ(frames[3].temporal_id, 2);
EXPECT_EQ(frames[4].temporal_id, 1);
EXPECT_EQ(frames[5].temporal_id, 2);
EXPECT_EQ(frames[6].temporal_id, 2);
EXPECT_TRUE(wrapper.FrameReferencesAreValid(frames));
}
TEST(ScalabilityStructureL3T3KeyTest,
ReenablingSpatialLayerBeforeMissedT0FrameDoesntTriggerAKeyFrame) {
ScalabilityStructureL3T3Key structure;
ScalabilityStructureWrapper wrapper(structure);
std::vector<GenericFrameInfo> frames;
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/2, /*s1=*/2));
wrapper.GenerateFrames(1, frames);
EXPECT_THAT(frames, SizeIs(2));
// Drop a spatial layer.
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/2, /*s1=*/0));
wrapper.GenerateFrames(1, frames);
EXPECT_THAT(frames, SizeIs(3));
// Reenable a spatial layer before T0 frame is encoded.
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/2, /*s1=*/2));
wrapper.GenerateFrames(1, frames);
EXPECT_THAT(frames, SizeIs(5));
EXPECT_EQ(frames[0].temporal_id, 0);
EXPECT_EQ(frames[1].temporal_id, 0);
EXPECT_EQ(frames[2].temporal_id, 1);
EXPECT_EQ(frames[3].temporal_id, 0);
EXPECT_EQ(frames[4].temporal_id, 0);
EXPECT_THAT(frames[3].frame_diffs, SizeIs(1));
EXPECT_THAT(frames[4].frame_diffs, SizeIs(1));
EXPECT_TRUE(wrapper.FrameReferencesAreValid(frames));
}
TEST(ScalabilityStructureL3T3KeyTest, ReenablingSpatialLayerTriggersKeyFrame) {
ScalabilityStructureL3T3Key structure;
ScalabilityStructureWrapper wrapper(structure);
std::vector<GenericFrameInfo> frames;
// Start with all spatial layers enabled.
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/2, /*s1=*/2, /*s2=*/2));
wrapper.GenerateFrames(3, frames);
EXPECT_THAT(frames, SizeIs(9));
// Drop a spatial layer. Two remaining spatial layers should just continue.
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/2, /*s1=*/0, /*s2=*/2));
wrapper.GenerateFrames(2, frames);
EXPECT_THAT(frames, SizeIs(13));
// Reenable spatial layer, expect a full restart.
structure.OnRatesUpdated(EnableTemporalLayers(/*s0=*/2, /*s1=*/2, /*s2=*/2));
wrapper.GenerateFrames(1, frames);
ASSERT_THAT(frames, SizeIs(16));
// First 3 temporal units with all spatial layers enabled.
EXPECT_EQ(frames[0].temporal_id, 0);
EXPECT_EQ(frames[3].temporal_id, 1);
EXPECT_EQ(frames[6].temporal_id, 0);
// 2 temporal units with spatial layer 1 disabled.
EXPECT_EQ(frames[9].spatial_id, 0);
EXPECT_EQ(frames[9].temporal_id, 1);
EXPECT_EQ(frames[10].spatial_id, 2);
EXPECT_EQ(frames[10].temporal_id, 1);
// T0 frames were encoded while spatial layer 1 is disabled.
EXPECT_EQ(frames[11].spatial_id, 0);
EXPECT_EQ(frames[11].temporal_id, 0);
EXPECT_EQ(frames[12].spatial_id, 2);
EXPECT_EQ(frames[12].temporal_id, 0);
// Key frame to reenable spatial layer 1.
EXPECT_THAT(frames[13].frame_diffs, IsEmpty());
EXPECT_THAT(frames[14].frame_diffs, ElementsAre(1));
EXPECT_THAT(frames[15].frame_diffs, ElementsAre(1));
EXPECT_EQ(frames[13].temporal_id, 0);
EXPECT_EQ(frames[14].temporal_id, 0);
EXPECT_EQ(frames[15].temporal_id, 0);
auto all_frames = rtc::MakeArrayView(frames.data(), frames.size());
EXPECT_TRUE(wrapper.FrameReferencesAreValid(all_frames.subview(0, 13)));
// Frames starting from the frame#13 should not reference any earlier frames.
EXPECT_TRUE(wrapper.FrameReferencesAreValid(all_frames.subview(13)));
}
} // namespace
} // namespace webrtc

View File

@ -1,102 +0,0 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "modules/video_coding/svc/scalability_structure_l2t1_key.h"
#include <utility>
#include <vector>
#include "absl/base/macros.h"
#include "api/transport/rtp/dependency_descriptor.h"
#include "rtc_base/checks.h"
#include "rtc_base/logging.h"
namespace webrtc {
namespace {
constexpr auto kNotPresent = DecodeTargetIndication::kNotPresent;
constexpr auto kSwitch = DecodeTargetIndication::kSwitch;
constexpr DecodeTargetIndication kDtis[3][2] = {
{kSwitch, kSwitch}, // Key, S0
{kSwitch, kNotPresent}, // Delta, S0
{kNotPresent, kSwitch}, // Key and Delta, S1
};
} // namespace
ScalabilityStructureL2T1Key::~ScalabilityStructureL2T1Key() = default;
ScalableVideoController::StreamLayersConfig
ScalabilityStructureL2T1Key::StreamConfig() const {
StreamLayersConfig result;
result.num_spatial_layers = 2;
result.num_temporal_layers = 1;
result.scaling_factor_num[0] = 1;
result.scaling_factor_den[0] = 2;
return result;
}
FrameDependencyStructure ScalabilityStructureL2T1Key::DependencyStructure()
const {
FrameDependencyStructure structure;
structure.num_decode_targets = 2;
structure.num_chains = 2;
structure.decode_target_protected_by_chain = {0, 1};
structure.templates.resize(4);
structure.templates[0].S(0).Dtis("S-").ChainDiffs({2, 1}).FrameDiffs({2});
structure.templates[1].S(0).Dtis("SS").ChainDiffs({0, 0});
structure.templates[2].S(1).Dtis("-S").ChainDiffs({1, 2}).FrameDiffs({2});
structure.templates[3].S(1).Dtis("-S").ChainDiffs({1, 1}).FrameDiffs({1});
return structure;
}
ScalableVideoController::LayerFrameConfig
ScalabilityStructureL2T1Key::KeyFrameConfig() const {
return LayerFrameConfig().Id(0).S(0).Keyframe().Update(0);
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureL2T1Key::NextFrameConfig(bool restart) {
std::vector<LayerFrameConfig> result(2);
// Buffer0 keeps latest S0T0 frame, Buffer1 keeps latest S1T0 frame.
if (restart || keyframe_) {
result[0] = KeyFrameConfig();
result[1].Id(2).S(1).Reference(0).Update(1);
keyframe_ = false;
} else {
result[0].Id(1).S(0).ReferenceAndUpdate(0);
result[1].Id(2).S(1).ReferenceAndUpdate(1);
}
return result;
}
GenericFrameInfo ScalabilityStructureL2T1Key::OnEncodeDone(
const LayerFrameConfig& config) {
RTC_CHECK_GE(config.Id(), 0);
RTC_CHECK_LT(config.Id(), ABSL_ARRAYSIZE(kDtis));
GenericFrameInfo frame_info;
frame_info.spatial_id = config.SpatialId();
frame_info.temporal_id = config.TemporalId();
frame_info.encoder_buffers = config.Buffers();
int config_id = config.IsKeyframe() ? 0 : config.Id();
frame_info.decode_target_indications.assign(std::begin(kDtis[config_id]),
std::end(kDtis[config_id]));
if (config.IsKeyframe()) {
frame_info.part_of_chain = {true, true};
} else {
frame_info.part_of_chain = {config.SpatialId() == 0,
config.SpatialId() == 1};
}
return frame_info;
}
} // namespace webrtc

View File

@ -1,42 +0,0 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_L2T1_KEY_H_
#define MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_L2T1_KEY_H_
#include <vector>
#include "api/transport/rtp/dependency_descriptor.h"
#include "common_video/generic_frame_descriptor/generic_frame_info.h"
#include "modules/video_coding/svc/scalable_video_controller.h"
namespace webrtc {
// S1 0--0--0-
// | ...
// S0 0--0--0-
class ScalabilityStructureL2T1Key : public ScalableVideoController {
public:
~ScalabilityStructureL2T1Key() override;
StreamLayersConfig StreamConfig() const override;
FrameDependencyStructure DependencyStructure() const override;
std::vector<LayerFrameConfig> NextFrameConfig(bool restart) override;
GenericFrameInfo OnEncodeDone(const LayerFrameConfig& config) override;
private:
LayerFrameConfig KeyFrameConfig() const;
bool keyframe_ = true;
};
} // namespace webrtc
#endif // MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_L2T1_KEY_H_

View File

@ -1,125 +0,0 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "modules/video_coding/svc/scalability_structure_l2t2_key.h"
#include <utility>
#include <vector>
#include "absl/base/macros.h"
#include "api/transport/rtp/dependency_descriptor.h"
#include "rtc_base/checks.h"
#include "rtc_base/logging.h"
namespace webrtc {
namespace {
constexpr auto kNotPresent = DecodeTargetIndication::kNotPresent;
constexpr auto kDiscardable = DecodeTargetIndication::kDiscardable;
constexpr auto kSwitch = DecodeTargetIndication::kSwitch;
// decode targets: S0T0, S0T1, S1T0, S1T1
constexpr DecodeTargetIndication kDtis[6][4] = {
{kSwitch, kSwitch, kSwitch, kSwitch}, // kKey, S0
{kNotPresent, kNotPresent, kSwitch, kSwitch}, // kKey, S1
{kNotPresent, kDiscardable, kNotPresent, kNotPresent}, // kDeltaT1, S0
{kNotPresent, kNotPresent, kNotPresent, kDiscardable}, // kDeltaT1, S1
{kSwitch, kSwitch, kNotPresent, kNotPresent}, // kDeltaT0, S0
{kNotPresent, kNotPresent, kSwitch, kSwitch}, // kDeltaT0, S1
};
} // namespace
ScalabilityStructureL2T2Key::~ScalabilityStructureL2T2Key() = default;
ScalableVideoController::StreamLayersConfig
ScalabilityStructureL2T2Key::StreamConfig() const {
StreamLayersConfig result;
result.num_spatial_layers = 2;
result.num_temporal_layers = 2;
result.scaling_factor_num[0] = 1;
result.scaling_factor_den[0] = 2;
return result;
}
FrameDependencyStructure ScalabilityStructureL2T2Key::DependencyStructure()
const {
FrameDependencyStructure structure;
structure.num_decode_targets = 4;
structure.num_chains = 2;
structure.decode_target_protected_by_chain = {0, 0, 1, 1};
structure.templates.resize(6);
auto& templates = structure.templates;
templates[0].S(0).T(0).Dtis("SSSS").ChainDiffs({0, 0});
templates[1].S(0).T(0).Dtis("SS--").ChainDiffs({4, 3}).FrameDiffs({4});
templates[2].S(0).T(1).Dtis("-D--").ChainDiffs({2, 1}).FrameDiffs({2});
templates[3].S(1).T(0).Dtis("--SS").ChainDiffs({1, 1}).FrameDiffs({1});
templates[4].S(1).T(0).Dtis("--SS").ChainDiffs({1, 4}).FrameDiffs({4});
templates[5].S(1).T(1).Dtis("---D").ChainDiffs({3, 2}).FrameDiffs({2});
return structure;
}
ScalableVideoController::LayerFrameConfig
ScalabilityStructureL2T2Key::KeyFrameConfig() const {
return LayerFrameConfig().Id(0).Keyframe().S(0).T(0).Update(0);
}
std::vector<ScalableVideoController::LayerFrameConfig>
ScalabilityStructureL2T2Key::NextFrameConfig(bool restart) {
if (restart) {
next_pattern_ = kKey;
}
std::vector<LayerFrameConfig> result(2);
// Buffer0 keeps latest S0T0 frame,
// Buffer1 keeps latest S1T0 frame.
switch (next_pattern_) {
case kKey:
result[0] = KeyFrameConfig();
result[1].Id(1).S(1).T(0).Reference(0).Update(1);
next_pattern_ = kDeltaT1;
break;
case kDeltaT1:
result[0].Id(2).S(0).T(1).Reference(0);
result[1].Id(3).S(1).T(1).Reference(1);
next_pattern_ = kDeltaT0;
break;
case kDeltaT0:
result[0].Id(4).S(0).T(0).ReferenceAndUpdate(0);
result[1].Id(5).S(1).T(0).ReferenceAndUpdate(1);
next_pattern_ = kDeltaT1;
break;
}
return result;
}
GenericFrameInfo ScalabilityStructureL2T2Key::OnEncodeDone(
const LayerFrameConfig& config) {
RTC_CHECK_GE(config.Id(), 0);
RTC_CHECK_LT(config.Id(), ABSL_ARRAYSIZE(kDtis));
GenericFrameInfo frame_info;
frame_info.spatial_id = config.SpatialId();
frame_info.temporal_id = config.TemporalId();
frame_info.encoder_buffers = config.Buffers();
int config_id = config.IsKeyframe() ? 0 : config.Id();
frame_info.decode_target_indications.assign(std::begin(kDtis[config_id]),
std::end(kDtis[config_id]));
if (config.IsKeyframe()) {
frame_info.part_of_chain = {true, true};
} else if (config.TemporalId() == 0) {
frame_info.part_of_chain = {config.SpatialId() == 0,
config.SpatialId() == 1};
} else {
frame_info.part_of_chain = {false, false};
}
return frame_info;
}
} // namespace webrtc

View File

@ -1,52 +0,0 @@
/*
* Copyright (c) 2020 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_L2T2_KEY_H_
#define MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_L2T2_KEY_H_
#include <vector>
#include "api/transport/rtp/dependency_descriptor.h"
#include "common_video/generic_frame_descriptor/generic_frame_info.h"
#include "modules/video_coding/svc/scalable_video_controller.h"
namespace webrtc {
// S1T1 0 0
// / / /
// S1T0 0---0---0
// | ...
// S0T1 | 0 0
// |/ / /
// S0T0 0---0---0
// Time-> 0 1 2 3 4
class ScalabilityStructureL2T2Key : public ScalableVideoController {
public:
~ScalabilityStructureL2T2Key() override;
StreamLayersConfig StreamConfig() const override;
FrameDependencyStructure DependencyStructure() const override;
std::vector<LayerFrameConfig> NextFrameConfig(bool restart) override;
GenericFrameInfo OnEncodeDone(const LayerFrameConfig& config) override;
private:
enum FramePattern {
kKey,
kDeltaT1,
kDeltaT0,
};
LayerFrameConfig KeyFrameConfig() const;
FramePattern next_pattern_ = kKey;
};
} // namespace webrtc
#endif // MODULES_VIDEO_CODING_SVC_SCALABILITY_STRUCTURE_L2T2_KEY_H_

View File

@ -14,6 +14,7 @@
#include <utility>
#include <vector>
#include "api/array_view.h"
#include "api/transport/rtp/dependency_descriptor.h"
#include "api/video/video_bitrate_allocation.h"
#include "api/video/video_frame_type.h"
@ -38,58 +39,66 @@ VideoBitrateAllocation EnableTemporalLayers(int s0, int s1, int s2) {
return bitrate;
}
std::vector<GenericFrameInfo> ScalabilityStructureWrapper::GenerateFrames(
void ScalabilityStructureWrapper::GenerateFrames(
int num_temporal_units,
bool restart) {
std::vector<GenericFrameInfo> frames;
std::vector<GenericFrameInfo>& frames) {
for (int i = 0; i < num_temporal_units; ++i) {
for (auto& layer_frame : structure_controller_.NextFrameConfig(restart)) {
for (auto& layer_frame :
structure_controller_.NextFrameConfig(/*restart=*/false)) {
int64_t frame_id = ++frame_id_;
bool is_keyframe = layer_frame.IsKeyframe();
absl::optional<GenericFrameInfo> frame_info =
structure_controller_.OnEncodeDone(std::move(layer_frame));
EXPECT_TRUE(frame_info.has_value());
GenericFrameInfo frame_info =
structure_controller_.OnEncodeDone(layer_frame);
if (is_keyframe) {
chain_diff_calculator_.Reset(frame_info->part_of_chain);
chain_diff_calculator_.Reset(frame_info.part_of_chain);
}
frame_info->chain_diffs =
chain_diff_calculator_.From(frame_id, frame_info->part_of_chain);
frame_info.chain_diffs =
chain_diff_calculator_.From(frame_id, frame_info.part_of_chain);
for (int64_t base_frame_id : frame_deps_calculator_.FromBuffersUsage(
is_keyframe ? VideoFrameType::kVideoFrameKey
: VideoFrameType::kVideoFrameDelta,
frame_id, frame_info->encoder_buffers)) {
EXPECT_LT(base_frame_id, frame_id);
EXPECT_GE(base_frame_id, 0);
frame_info->frame_diffs.push_back(frame_id - base_frame_id);
frame_id, frame_info.encoder_buffers)) {
frame_info.frame_diffs.push_back(frame_id - base_frame_id);
}
frames.push_back(*std::move(frame_info));
frames.push_back(std::move(frame_info));
}
restart = false;
}
}
if (restart) {
buffer_contains_frame_.reset();
}
for (const GenericFrameInfo& frame : frames) {
bool ScalabilityStructureWrapper::FrameReferencesAreValid(
rtc::ArrayView<const GenericFrameInfo> frames) const {
bool valid = true;
// VP9 and AV1 supports up to 8 buffers. Expect no more buffers are not used.
std::bitset<8> buffer_contains_frame;
for (size_t i = 0; i < frames.size(); ++i) {
const GenericFrameInfo& frame = frames[i];
for (const CodecBufferUsage& buffer_usage : frame.encoder_buffers) {
if (buffer_usage.id < 0 || buffer_usage.id >= 8) {
ADD_FAILURE() << "Invalid buffer id " << buffer_usage.id
<< " for frame#" << i
<< ". Up to 8 buffers are supported.";
valid = false;
continue;
}
if (buffer_usage.referenced && !buffer_contains_frame_[buffer_usage.id]) {
ADD_FAILURE() << "buffer " << buffer_usage.id
if (buffer_usage.referenced && !buffer_contains_frame[buffer_usage.id]) {
ADD_FAILURE() << "buffer " << buffer_usage.id << " for frame#" << i
<< " was reference before updated.";
valid = false;
}
if (buffer_usage.updated) {
buffer_contains_frame_.set(buffer_usage.id);
buffer_contains_frame.set(buffer_usage.id);
}
}
for (int fdiff : frame.frame_diffs) {
if (fdiff <= 0 || static_cast<size_t>(fdiff) > i) {
ADD_FAILURE() << "Invalid frame diff " << fdiff << " for frame#" << i;
valid = false;
}
}
}
return frames;
return valid;
}
} // namespace webrtc

View File

@ -14,6 +14,7 @@
#include <vector>
#include "api/array_view.h"
#include "api/transport/rtp/dependency_descriptor.h"
#include "api/video/video_bitrate_allocation.h"
#include "common_video/generic_frame_descriptor/generic_frame_info.h"
@ -32,16 +33,22 @@ class ScalabilityStructureWrapper {
explicit ScalabilityStructureWrapper(ScalableVideoController& structure)
: structure_controller_(structure) {}
std::vector<GenericFrameInfo> GenerateFrames(int num_tempral_units,
bool restart);
std::vector<GenericFrameInfo> GenerateFrames(int num_temporal_units) {
return GenerateFrames(num_temporal_units, /*restart=*/false);
std::vector<GenericFrameInfo> frames;
GenerateFrames(num_temporal_units, frames);
return frames;
}
void GenerateFrames(int num_temporal_units,
std::vector<GenericFrameInfo>& frames);
// Returns false and ADD_FAILUREs for frames with invalid references.
// In particular validates no frame frame reference to frame before frames[0].
// In error messages frames are indexed starting with 0.
bool FrameReferencesAreValid(
rtc::ArrayView<const GenericFrameInfo> frames) const;
private:
ScalableVideoController& structure_controller_;
std::bitset<8> buffer_contains_frame_ = 0;
FrameDependenciesCalculator frame_deps_calculator_;
ChainDiffCalculator chain_diff_calculator_;
int64_t frame_id_ = 0;

View File

@ -305,27 +305,29 @@ INSTANTIATE_TEST_SUITE_P(
SvcTestParam{"S2T1", /*num_temporal_units=*/3},
SvcTestParam{"L2T2", /*num_temporal_units=*/4},
SvcTestParam{"L2T2_KEY", /*num_temporal_units=*/4},
SvcTestParam{"L2T2_KEY_SHIFT", /*num_temporal_units=*/4}),
SvcTestParam{"L2T2_KEY_SHIFT", /*num_temporal_units=*/4},
SvcTestParam{"L3T3_KEY", /*num_temporal_units=*/8}),
[](const testing::TestParamInfo<SvcTestParam>& info) {
return info.param.name;
});
// TODO(danilchap): Merge with ScalabilityStructureTest when the functionality
// is implemented for all tested structures.
INSTANTIATE_TEST_SUITE_P(Svc,
ScalabilityStructureSetRatesTest,
Values(SvcTestParam{"L1T2",
/*num_temporal_units=*/4},
SvcTestParam{"L1T3", /*num_temporal_units=*/8},
SvcTestParam{"L2T1",
/*num_temporal_units=*/3},
SvcTestParam{"L2T2",
/*num_temporal_units=*/4},
SvcTestParam{"L3T1", /*num_temporal_units=*/3},
SvcTestParam{"L3T3", /*num_temporal_units=*/8}),
[](const testing::TestParamInfo<SvcTestParam>& info) {
return info.param.name;
});
INSTANTIATE_TEST_SUITE_P(
Svc,
ScalabilityStructureSetRatesTest,
Values(SvcTestParam{"L1T2", /*num_temporal_units=*/4},
SvcTestParam{"L1T3", /*num_temporal_units=*/8},
SvcTestParam{"L2T1", /*num_temporal_units=*/3},
SvcTestParam{"L2T1_KEY", /*num_temporal_units=*/3},
SvcTestParam{"L2T2", /*num_temporal_units=*/4},
SvcTestParam{"L2T2_KEY", /*num_temporal_units=*/4},
SvcTestParam{"L3T1", /*num_temporal_units=*/3},
SvcTestParam{"L3T3", /*num_temporal_units=*/8},
SvcTestParam{"L3T3_KEY", /*num_temporal_units=*/8}),
[](const testing::TestParamInfo<SvcTestParam>& info) {
return info.param.name;
});
} // namespace
} // namespace webrtc