Current thresholds were tuned to guarantee no buffer overshoot in an extreme scenario (encoding a high complexity video in a low bitrate).
Bug: b/337757868, webrtc:351644568
Change-Id: I832b2564af6f18f06550338cc9b3618f8acdf831
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/356580
Reviewed-by: Dan Tan <dwtan@google.com>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#42620}
For better consistency with the rest codebase (it is min_/max_ for all params in video_encoder.h; only qp is for some reason prefixed with minimum_).
Also fixed constant names in libaom AV1 encoder wrapper (moved min from suffix to prefix, minimum -> min_).
Bug: chromium:328598314
Change-Id: I6d8521a3abff3a0595a5241c02ef4746eb4694df
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/356600
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#42604}
Before this change the AV1 encoder wrapper converted target frame rate from double to integer with rounding to the middle. That approach resulted in a bitrate mismatch caused by rounding error. The mismatch was especially high at low frame rates. For example, at target frame rate 1.4fps the bitrate mismatch reached 40%:
out/debug/video_codec_perf_tests --gtest_also_run_disabled_tests --gtest_filter=*EncodeDecode --framerate_fps=1.4 --width=320 --height=180 --bitrate_kbps=32 --num_frames=600
...
RESULT s0t0_bitrate_mismatch_pct: DISABLED_EncodeDecode= {39.171875,0} n%
After the change the mismatch reduced to ~2% in the same scenario:
RESULT s0t0_bitrate_mismatch_pct: DISABLED_EncodeDecode= {-2.178125,0} n%
Bug: b/337757868
Change-Id: Ia51f92b3dfdce103eed1d04cac0e084b69fa8213
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/356500
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#42601}
The minimum QP field will be used to signal what the QP value will be
once the encoder reach its target video quality. This will be used
in the generalized QP convergence detection.
Bug: chromium:328598314
Change-Id: I82299cd921e3c091e651218d1e3f337875176567
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/355701
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Johannes Kron <kron@webrtc.org>
Reviewed-by: Markus Handell <handellm@google.com>
Cr-Commit-Position: refs/heads/main@{#42559}
Set layer QP limits equal to QP limits in the main encoder config. This reduces number of nodes to modify if you need to change the settings.
Bug: b/337757868
Change-Id: Id7f6f9d6527903e8e22ff4fad2c974bee6e87cb3
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/353982
Reviewed-by: Marco Paniconi <marpan@webrtc.org>
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Reviewed-by: Sergey Silkin <ssilkin@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#42466}
This was a kill-switch for frame dropping in AV1 encoder. The frame dropping was enabled in June 2023. Since we have not heard about about any issues related to the frame dropping, we can remove the field trial.
Bug: webrtc:42225542
Change-Id: I4b2f1d5ff61e4ae3a4a7fc6711bb83f7d522fc6a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/349921
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Reviewed-by: Åsa Persson <asapersson@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#42241}
Instead of from the global field trial string.
Bug: webrtc:42220378
Change-Id: Iddb41429e388792de02f702b4caa35689c57d9e5
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/347720
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#42201}
The width and height of mapped_buffer must match the d_w and d_h members
of frame_to_encode_, which is passed to aom_codec_encode().
Bug: b:330482827
Change-Id: I85d8c82133768685565f165eafc893c42dc40b12
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/345807
Reviewed-by: Sergey Silkin <ssilkin@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Wan-Teh Chang <wtc@google.com>
Cr-Commit-Position: refs/heads/main@{#42036}
There's an AV1 encoder speed setting 11 that is supposed to be used
for screen sharing content.
Bug: chromium:328598314
Change-Id: Id97898554a740eb1684d03c782c718c19f4c95e2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/342201
Commit-Queue: Johannes Kron <kron@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41874}
ARM-specific settings were intended to be used on mobile ARM devices which may not be powerful enough. But the settings were also applied to ARM-based Macs. This changes restricts ARM-specific settings to Android and iOS platforms.
Bug: none
Change-Id: I68764b4c0679db07399bba5923f4a6be89c5ad80
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/321861
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Reviewed-by: Jerome Jiang <jianj@google.com>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40884}
To avoid name collision with Timestamp type,
To avoid confusion with capture time represented as Timestamp
Bug: webrtc:9378
Change-Id: I8438a9cf4316e5f81d98c2af9dc9454c21c78e70
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/320601
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Reviewed-by: Rasmus Brandt <brandtr@webrtc.org>
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40796}
The main goal of this change is to disable the quality scaler when multiple spatial layers are used.
Bug: b/295129711
Change-Id: I25e0b7440a8c2adee3e97720a1e0ee5e0a914334
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/319181
Reviewed-by: Sergey Silkin <ssilkin@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Philip Eliasson <philipel@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40709}
Use 8 threads for > 720p
Use 4 tile columns and 2 tile rows for 8 threads
Use 2 tile columns and 2 tile rows for 4 threads
For VGA, 2 tile_col x 2 tile_row configuration has
- ~9% speedup over 4 tile_col x 1 tile_row
- ~5% speedup over 1 tile_col x 4 tile_row
Bug: None
Change-Id: I3c1ea948437aece7c6734ce25351158cbdf0a15b
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/307880
Commit-Queue: Jerome Jiang <jianj@google.com>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40237}
Error resilience is no longer required for upper temporal layers.
Disabling error resilience on the upper layers leads to a ~2% PSNR BD-rate gain.
Reland of https://webrtc-review.googlesource.com/c/src/+/302001
Bug: webrtc:15106
Change-Id: I72ca9d504a7848dda934cbd52669027061742256
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/305782
Reviewed-by: Jerome Jiang <jianj@google.com>
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Marco Paniconi <marpan@webrtc.org>
Reviewed-by: Michael Horowitz <mhoro@webrtc.org>
Reviewed-by: Marco Paniconi <marpan@google.com>
Cr-Commit-Position: refs/heads/main@{#40099}
This reverts commit 2080dacfb7.
Reason for revert: This CL is causing a lot of flakiness on iOS bots
https://ci.chromium.org/p/webrtc/builders/ci/iOS%20Debug%20%28simulator%29
Original change's description:
> For AV1, disable error resilience on upper temporal layers. Error resilience is no longer required for upper temporal layers. Disabling error resilience on the upper layers leads to a ~2% PSNR BD-rate gain.
>
> Bug: webrtc:15106
> Change-Id: Id92d51defbd26c1a77e3c9fe19607e9db4a3e7c1
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/302001
> Reviewed-by: Marco Paniconi <marpan@webrtc.org>
> Commit-Queue: Michael Horowitz <mhoro@webrtc.org>
> Cr-Commit-Position: refs/heads/main@{#39900}
Bug: webrtc:15106
Change-Id: I24515280113ed6681c9766026ec24d689035c031
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/301983
Bot-Commit: rubber-stamper@appspot.gserviceaccount.com <rubber-stamper@appspot.gserviceaccount.com>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Commit-Queue: Jeremy Leconte <jleconte@google.com>
Owners-Override: Jeremy Leconte <jleconte@google.com>
Cr-Commit-Position: refs/heads/main@{#39903}
Fix Issue 15059: The target bitrate was mistakenly set to be the maximal
bitrate when initializing the libaom encoder.
Bug: webrtc:15059
Change-Id: I38498d4cce7b0a9c26736d9f1096178dd2e1fef6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/300004
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39822}
libaom uses the quantizer as an index for an array of size 64, so
encoder_settings_.qpMax must be <= 63.
Add a comment to LibaomAv1Encoder::SetSvcParams() to explain why the
method doesn't initialize svc_params.layer_target_bitrate.
Bug: None
Change-Id: I26be80de005752214365abbe8b9b32dc976cee0b
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/293680
Reviewed-by: Erik Språng <sprang@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Commit-Queue: Wan-Teh Chang <wtc@google.com>
Cr-Commit-Position: refs/heads/main@{#39572}
This can happen when the encoder uses real presentation timestamps that
originate with the input frames. By using those, the encoder can bypass
webrtc frame dropping logic and may severely over/under-shoot if the
timestamps are very precise. In practice, this seems rather common on
Chrome on Windows.
Bug: aomedia:3391
Change-Id: I2be5eed4fabc86dac8a6c7bfdd068c2dcb5a3743
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/294740
Commit-Queue: Erik Språng <sprang@webrtc.org>
Reviewed-by: Philip Eliasson <philipel@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39382}
This CL propagates capture_time_identifier introduced in
webrtc::VideoFrame and propagates it to EncodedImage. For use cases
involving EncodedTransforms, this identifier is further propagated to
TransformableVideoSenderFrame.
VideoEncoder::Encode function is overriden by each encoder. Each of
these overriden functions needs to be changed so that they can handle
this new identifier and propagate its value in the created EncodedImage.
Change-Id: I5bea4c5a3fe714f1198e497a4bcb5fd059afe516
Bug: webrtc:14878
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/291800
Reviewed-by: Tony Herre <herre@google.com>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Commit-Queue: Palak Agarwal <agpalak@google.com>
Cr-Commit-Position: refs/heads/main@{#39374}
The kMinimumFrameRate constant is only used in a comparison with
RateControlParameters::framerate_fps, which is of the double type.
Declare kMinimumFrameRate as double to match.
Note: The kMinimumFrameRate constant was added in
https://webrtc-review.googlesource.com/c/src/+/170360.
Bug: webrtc:11404
Change-Id: I11769867d4e52a720219c8a0ade8e8b74d13ca86
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/293384
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Wan-Teh Chang <wtc@google.com>
Cr-Commit-Position: refs/heads/main@{#39320}
The elements of the fps_allocation vector are fractions of the maximum
frame rate. Each fraction is represented as an 8-bit unsigned integer,
where 0 = 0% and 255 = 100%.
The original code (added in
https://webrtc-review.googlesource.com/c/src/+/201384) sets the elements
of the fps_allocation vector to frame rates rather than frame rate
fractions. Perhaps fps_allocation could be renamed to avoid this kind of
confusion.
modules_unittests --gtest_filter=LibaomAv1EncoderTest.*
Tested:
Change-Id: Icd050da3b3c2cff31913c3430f7b6b6e9829b9fa
Bug: None
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/292784
Commit-Queue: Wan-Teh Chang <wtc@google.com>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Reviewed-by: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39286}
This is in prep for the migration of all unwrappers to
SequenceNumberUnwrapper as a standard implementation.
This moves the SeqNumUnwapper to its own header and adds 2 methods to
SeqNumUnwrapper which are defined by other unwrappers:
* PeekUnwrap
* Reset
It also adds two implementations for RtpTimestamps and
RtpSequenceNumbers.
Bug: webrtc:13982
Change-Id: I5baefb2de1db92fe1bb600760bd63b71e9310eb5
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/288742
Commit-Queue: Evan Shrubsole <eshr@webrtc.org>
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Auto-Submit: Evan Shrubsole <eshr@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39030}
This is in the webrtc-stats spec at
https://www.w3.org/TR/webrtc-stats/#dom-rtcoutboundrtpstreamstats-scalabilitymode.
This adds the scalability mode to CodecSpecificInfo which is used to
plumb the modes for each simulcast layer.
TBR=orphis@webrtc.org
Tested: Compiled into Chrome and confirmed the scalability mode set for AV1, VP9, VP8 and H264 software encoders in chrome://webrtc-internals.
Bug: webrtc:14730
Change-Id: I71ceba8f6485a4f4a73e0856031b8d5f16f913f2
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/285085
Reviewed-by: Harald Alvestrand <hta@webrtc.org>
Reviewed-by: Åsa Persson <asapersson@webrtc.org>
Commit-Queue: Evan Shrubsole <eshr@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#38847}
The tests check that the various scalability mode are supported
and the frames are marked properly by the encoder with their
spatial and temporal index.
The same information is then checked on the receiving side.
A new member is added on EncodedImage to store the temporal index,
and is filled by the encoders and retreived by the ref finder
objects on the decoding side.
Bug: webrtc:11607
Change-Id: I7522f6a6fc5402244cab0c4c64b544ce09bc5204
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/260189
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Rasmus Brandt <brandtr@webrtc.org>
Reviewed-by: Artem Titov <titovartem@webrtc.org>
Commit-Queue: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#37303}
Currently `CreateLibaomAv1Encoder` will either return an actual libaom AV1 encoder or a nullptr depening on whether the build flag `enable_libaom` was configured to true or not. This CL updates the `libaom_av1_encoder` build target to no longer depend on `enable_libaom` so that `CreateLibaomAv1Encoder` will always return an encoder instance.
Added `CreateLibaomAv1EncoderIfSupported` as a replacement to the old `CreateLibaomAv1Encoder`.
Bug: webrtc:13573
Change-Id: Ibdcd52c609acd79feefa2b86f19d1b4ca3e91d0a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/242360
Reviewed-by: Erik Språng <sprang@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Peter Hanspers <peterhanspers@webrtc.org>
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Commit-Queue: Philip Eliasson <philipel@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#35763}
