Plumbing hdrMode through the default asset loader factory via the constructor is problematic because it breaks API boundaries. It means there is another way to set hdrMode outside of Composition.java and TransformationRequest.java, which is error prone and cause problems if someone an app starts customizing the assetloaderfactory. It also means custom asset loaders can't receive this information without hacking around.
The introduction of the composition-level settings class makes this approach easily extensible for other settings applied on the composition level but use in an individual asset level basis (e.g. ultraHDR support).
PiperOrigin-RevId: 611466920
Previously, the track format was used in VideoSampleExporter. Now, we use a
simulated decoder output format.
As the last expected change for this bug, also adds release notes
PiperOrigin-RevId: 609080629
MediaCodec docs already allude to potentially mismatching H.264 level
between container and bitstream. Relax the initialization data check to
reflect this.
PiperOrigin-RevId: 608942322
The new API will take both `creation time` and `modification time`.
Till now, Mp4Muxer wrote `modification time` in both
`creation time` and `modification time` field, which was
incorrect.
PiperOrigin-RevId: 605590623
Change the file extension retrieval back to how it was before 5488d33da8 to reduce the change of false negatives in `isImage()`
PiperOrigin-RevId: 605281186
Before supporting transmuxing when both no op effects and regular rotations are set, move setting the muxerWrapper rotation out of shouldTranscodeVideo() to ensure the muxerWrapper rotation is only set at the appropriate times.
This cl also ensures the state between the muxerWrapper and the list of video effects is consistent by clearing the list of videoEffects in trim optimization. If trim optimisation is being applied, then EditedMediItem.effects.videoEffects only contains no-op effects or regular rotations that get be applied in the muxer wrapper. Therefore, we should clear the list of video effects to ensure that no effect gets applied twice.
PiperOrigin-RevId: 604292052
Transformer export and ExoPlayer previewing both read inputColorInfo from
registerInputStream now, instead of maintaining a consistent input color
throughout multiple streams in a sequence.
Therefore, we can remove inputColor-related arguments and methods now.
PiperOrigin-RevId: 603423509
The `MediaMuxer` (FrameworkMuxer) supports AV1 mime type from API 34.
For this to work track `Format/MediaFormat` must have `CSD`
data.
Change also include providing AV1 option in demo app.
Transmuxing of an AV1 mime type input will not work because
`Extractor` does not extract AV1 CSD data.
Verified changes manually via demo app.
PiperOrigin-RevId: 603002380
Previously, input assets had to be all SDR or all HDR.
After this CL, if tone-mapping is requested, HDR and SDR may mix in any order. If tone-mapping is not requested, SDR may precede HDR, but not vice versa, until SDR to HDR tone-mapping is implemented
Some changes to accomplish this include:
1. Inputting the decoded format's color to VideoFrameProcessor.registerInputStream
for each stream.
2. Calculating the estimated decoded format's color for each stream, by estimating
it based on MediaCodec tone-mapping.
PiperOrigin-RevId: 602747837
If trim optimisation is being applied, then `EditedMediItem.effects.videoEffects` only contains no-op effects or regular rotations that get be applied in the muxer wrapper. Therefore, we should clear the list of video effects to ensure that no effect gets applied twice.
This fixes requested rotations being applied twice in trim optimization.
Manually tested to ensure all combinations of trimming+muting+rotating works at intended.
PiperOrigin-RevId: 601853203
Many usages are needed to support other deprecations and some
can be replaced by the recommended direct alternative.
Also replace links to deprecated/redirected dev site
PiperOrigin-RevId: 601795998
since we now can support taking in the inputColor upon registering the stream, there is no need to hardcode the image input color anymore. We should now be able to support switching between texture and image input which we couldn't before, but this is untested and not necessary.
PiperOrigin-RevId: 601784149
This allows us to also plumb whether we're doing MediaCodec tone-mapping,
which will be used in a follow-up CL in the ExoAssetLoaderVideoRenderer.
PiperOrigin-RevId: 601774435
Allow setAdditionalRotationDegrees to be called with same rotation after tracks added. This is needed for processes that mux files partially trim optimization so they don't error out after hitting the check state
Manually tested to ensure trim optimization succeeds, automated test added here as well
PiperOrigin-RevId: 601081778
[Android best
practices](https://developer.android.com/media/optimize/sharing#android_8_81_and_9)
recommend disabling B-frames on API 27, but some devices output B-frames anyway
when H.264/AVC High profile is selected. Add a workaround to force these
devices not to use high profile, to prevent B-frames being output.
`MediaMuxer` likely handles B-frames on these specific devices, but this change
allows the migration to default to in-app muxing to take place without
introducing errors, and it's a temporary workaround until B-frames are properly
supported in the in-app muxer.
PiperOrigin-RevId: 600422238
Although not public, documenting what happens in each state allows for
better code understanding at a glance.
As part of this, refactored the #getProgress method to highlight that
the other options are "non standard".
PiperOrigin-RevId: 599505369
Moves setting bitmapFactory options from BitmapLoader to DatasourceBitmapLoader
BitmapLoader is a general interface for bitmap loading that could use loading implementations other that BitmapFactory, with the rise of Glide being a loader of choice. It's best to correct this interface so that it remains generic
We can't deprecate easily because the other loadBitmap method in that case has a default implementation that relies on the first one, so the change is still breaking. BitmapLoader is public api in common, but it's @UnstableAPI and hasn't been around for very long (be38670391 added it), so it seems this is the best way forward.
PiperOrigin-RevId: 597897098
Replace the event for notifying fallback to cover codec initialization in
general (but keeping the list of errors).
Add a flag to control whether to try non-primary codecs, with the same
documentation as the similar flag in ExoPlayer's renderer.
Make the class final as it shouldn't be necessary to subclass it.
PiperOrigin-RevId: 592869868
Move `inputColorInfo` from `VideoFrameProcessor`'s `Factory.create` to `FrameInfo`,
input via `registerInputStream`.
Also, manually tested on exoplayer demo that setVideoEffects still works.
PiperOrigin-RevId: 591273545
Also adds an alternate way to configure the AudioGraph.
Apps should no longer need to ensure that inputs have the same sample
rate.
PiperOrigin-RevId: 588747431
Added a new ABANDONED option so one can tell the difference between when the optimization has been requested but not applied vs not requested at all. also changed the ordering do better represent the hierarchy of failure modes
PiperOrigin-RevId: 588720513
The issue that motivated adding this (frames unexpectedly being dropped by the
decoder) has been addressed, so we can turn off the logging to reduce
unnecessary allocations during transformation. We can easily turn on debug
logging in future as needed by setting `DebugTraceUtil.DEBUG = true`.
Also avoid allocations for string building at logging call sites by passing a
format string for extra info. Formatting the string now only happens when
debugging is turned on.
Tested manually by running transformations in the new state (DEBUG = false) and
with debugging turned on.
PiperOrigin-RevId: 585666349
The decoder and encoder won't accept high values for frame rate, so avoid
setting the key when configuring the decoder, and set a default value for the
encoder (where the key is required).
Also skip SSIM calculation for 4k, where the device lacks concurrent decoding
support.
PiperOrigin-RevId: 585604976
Using `Integer.MAX_VALUE` risks causing arithmetic overflow in the codec
implementation.
Issue: androidx/media#810
#minor-release
PiperOrigin-RevId: 585104621
These usages have no need for the double ended input functionality. All
other usages across media3 are ConcurrentLinkedQueue.
PiperOrigin-RevId: 584841104
Composition and EditedMediaItemSequence don't allow empty lists in their main
constructors, so shouldn't the vararg API. This is more inline with Effective
Java item 53.
PiperOrigin-RevId: 583415124
When transmuxing, the `EncodedSampleExporter` maintains a queue of input
buffers that get filled with encoded data by the asset loader. The number of
buffers was limited to avoid using more and more memory if producer (asset
loader) gets far ahead of the consumer (exporter).
Previously this limit was fixed at 10 buffers, but increasing the number of
buffers can make some transmux operations much faster. Allow allocating between
a min and max number of buffers, and also set a target allocation size beyond
which new buffers can't be allocated. This allows audio formats which require
many small buffers to be processed more quickly, while preventing allocating
too much memory for hypothetical very high bitrate formats.
'Remove video' edits on local videos in particular get much faster, because
audio buffers are very short and there are lots of them. With a sample 10
minute video, a 'remove video' edit took 2 seconds (36 seconds before this
change). With a sample 1 minute removing video took 0.25 seconds after this
change (2.5 seconds before).
The speed improvement is smaller for other types of edits that retain the video
track. Transmuxing a 10 minute video retaining the video track took 26 seconds
(40 seconds before).
PiperOrigin-RevId: 583390284
MIME types are case-insensitive, but none of the many existing
comparisons across our code base take this into account. The
code can be made more robust by normalizing all MIME types at the
moment they are first set into a class/builder and adding toLowerCase
as part of the normalization.
Most concretely, this fixes an issue with playing HLS streams via
the IMA SDK where the stream MIME type is indicated with all lower
case "application/x-mpegurl", which failed the MIME type comparison
in DefaultMediaSourceFactory.
PiperOrigin-RevId: 582317261
The dump file diff as part of this change is because using AudioGraph
means the 2nd item is automatically edited to match the AudioFormat of
the 1st item {44.1KHz mono}, rather than {48KHz stereo}.
Manually verified that for the 2nd item, data output:
* Before: 66936 bytes (16734 frames) output = 348_625us of audio.
* After: 30750 bytes (15375 frames) output = 348_639us of audio.
The small final buffer is caused by SonicAudioProcessor outputting all
pending data when EOS queued, and is WAI.
PiperOrigin-RevId: 580494578
Changes includes:
1. Add resume flow.
2. Change demo app code to resume export.
3. Changes in test infra to trigger resume.
4. E2E Test cases
PiperOrigin-RevId: 579895744
When transmuxing, we usually only need to offset the timestamp by the position of a mediaItem in a sequence.
Trim optimization introduces another type of offset: for the transmux of the second part of the video we need to offset the timestamps by the total duration already trancoded by transformer.
PiperOrigin-RevId: 576134656
MediaCodecRenderer has already been updated to not rely on the
input stream to mark its samples as decode-only and instead use
a simple time-based comparison to achieve the same effect.
This change makes the same update for all other renderers that
either use the flag directly or forward to a "decoder" instance.
PiperOrigin-RevId: 568232212
If getProgress is blocking whilst the internal thread calls endInternal
(for error or success), the condition is never opened. Related to this,
onCompleted and onError are therefore never surfaced to the app.
progressState is accessed from application and internal threads, so
should be marked volatile to prevent a thread caching the value.
PiperOrigin-RevId: 565720184
When we switch from MUXER_MODE_MUX_PARTIAL_VIDEO to MUXER_MODE_APPEND_VIDEO
`muxedPartialVideo` will already be `true` so `endTrack` method will pass
through this `if(muxedPartialVideo)` check which is incorrect.
PiperOrigin-RevId: 565398117
The logic that handles components' boundaries are grouped together in private
methods, like handling VideoCompositor's output textures.
PiperOrigin-RevId: 565131579
For pause and resume feature we need to use same `MuxerWrapper`
for `remuxing processed video` and then to `process remaining video`.
In order to use same `MuxerWrapper` across `different Exports`
we need to preserve its state.
PiperOrigin-RevId: 564728396
For pause and resume feature we will need to build `ExportResult`
from multiple internal Exports. Keeping the `ExportResultBuider` in
`Transformer` class will allow using same builder across different
internal exports.
PiperOrigin-RevId: 563392443
For pause and resume feature we will remux the previously processed
video in the first export and then in the second export we will resume
processing remaining video. For the second export we will have to
set initial timestamp offset so that video samples from the second
export are in continuation to the previously muxed samples.
PiperOrigin-RevId: 561689651
For pause and resume we need to perform multiple intermediate exports.
So moved start() logic into a separate so that it can be reused from
different places with different Compositions and MuxerWrapper.
PiperOrigin-RevId: 561675472
For pause/resume feature, same MuxerWrapper needs to be
used across intermediate exports. So pass the MuxerWrapper
from Transformer.java
More specifically, when resume() is called
1. Create a MuxerWrapper and remux the previous
video samples (Export 1).
2. User the same MuxerWrapper and start processing
remaining video samples (Export 2).
PiperOrigin-RevId: 561325867
More specifically, this CL
- Defines a VideoGraph interface
- Adds a factory method, to hide the constructors
- Separate out an initialize method that does the real work (following that constructor should not do real work)
- The VideoGraph takes in a list of composition effects. For now, we only use the `Presentation`, if there is one in the list. This means we can take any number of Presentations now, but only the first one will be used.
- Moves the VideoFrameProcessingWrapper to its own file
PiperOrigin-RevId: 561059653
For pause and resume feature we need to find the timestamp of
the last sync sample in an MP4 file. The Mp4ExtractorWrapper
provides an easy to use API on top of Mp4Extractor.
PiperOrigin-RevId: 560113112
The first call to method `registerInputStream` doesn't block.
Later successive calls to the method blocks until the previous register call
finishes.
PiperOrigin-RevId: 559694490
We will need to build upon original composition
for pause and resume feature and will have to set modified sequences.
In the next CL, will add method to set sequences on builder and will
probably remove it from the constructor parameter (to avoid providing
two APIs for same thing).
PiperOrigin-RevId: 558127334
I wasn't able to reproduce the error in the bug by running videoTranscoding_withTextureInput_completesWithCorrectFrameCountAndDuration
100 times, but I figured this added debug info may help the next time the
issue is seen.
PiperOrigin-RevId: 558118443
Rename AudioMixerImpl to DefaultAudioMixer.
Removes the AudioMixerImpl specific getOutputAudioFormat, as the caller
defines and sets this.
PiperOrigin-RevId: 555887722
ExoPlayer queues the EOS buffer to the decoder with offset/size/timestamp all
equal to zero, and a EOS flag.
69769c77b3 set TIME_END_OF_SOURCE on the EOS buffer from the extractor.
Queueing the EOS buffer to the decoder with TIME_END_OF_SOURCE causes some
decoders to output wrong timestamps in its output.
This CL replicates what ExoPlayer does in Transformer.
PiperOrigin-RevId: 553104213
After this change, every queued bitmap is treated as an individual input stream
(like a new MediaItems).
This change merges the FrameDropTest and FrameDropPixelTest into one (while maintaining all the test cases)
- This is accomplished by generating bitmaps with timestamps on it in FrameDropTest and compare them with goldens (one may call this a pixel test, please lmk if you want this to be renamed)
- The most part of the change comes from DefaultVideoFrameProcessorVideoFrameRenderingTest. The overall working is
- We bypass the input manager
- The TestFrameGenerator generates frames based on timestamps. In this case, we generate frames with timestamps on it
- The generated frame is sent to texture output and in turn saved to bitmaps
- We then compare the generated bitmap with the goldens
PiperOrigin-RevId: 551795770
This should make no functional difference because `SampleExporter` always
checks for H.265 and H.264 first. However, in case we ever change that code,
these are used in priority order so it's better to order them accordingly.
PiperOrigin-RevId: 550894935
When generating silence for AudioProcessingPipeline, audio never
queued EOS downstream.
Linked to this, when silence followed an item with audio, the silence
was added to SilentAudioGenerator before the mediaItem reconfiguration
occurred. If the silence had effects, the APP would be flushed after
silence queued EOS, resetting APP.isEnded back to false, so AudioGraph
never ended.
Regression tests reproduce failure without fix, but pass with it.
PiperOrigin-RevId: 550853714
For each event, the timestamp and presentation time is logged. The trace can
then be dumped to a tsv file and easily imported in a spreadsheet.
PiperOrigin-RevId: 550839156
AudioMixingUtil#mix handles input & output in float or Int16 PCM. Given
Float and Int16 use different sample ratnes, this util handles
conversion between the two, based on the encoding being mixed to.
Migrate AudioMixer to use the util, removing AudioMixingAlgorithm
interface and implementation. ChannelMixingAudioProcessor will be
migrated after additional performance checks.
PiperOrigin-RevId: 548994584
SingleInputVideoGraph implements GraphInput now, so the asset loaders would
interface directly with SIVG, rather than VideoSampleExporter. This is to pave
way for multi-asset video processing.
PiperOrigin-RevId: 547561042
durationUs is almost always going to be a larger number than the sample
rate, so pass as the main value, rather than the multiplier.
PiperOrigin-RevId: 547193927
Add documentation for threading requirements at the class level (in
addition to existing documentation on the methods) to improve
discoverablility. Also fix a couple of nits in the javadoc (US English
spelling, avoid passive voice) and in `OnInputFrameProcessedListener`.
PiperOrigin-RevId: 546303732
On a MediaItem change, the input Format (and Effects to apply) may be
different. Therefore the AudioProcessingPipeline must be reconfigured
to determine what processing is active, and what the AudioFormat of the
data output is. In the event that it is different, additional
AudioProcessor instances must be used to ensure the encoder will still
be able to accept the audio buffers.
PiperOrigin-RevId: 544338451
This fixes a broken javadoc link and makes it possible to customize media
source creation via the asset loader factory without also having to
implementing a decoder factory from scratch.
PiperOrigin-RevId: 543995538
MP4 edit lists sometimes ask to start playback between two samples.
If this happens, we currently change the timestamp of the first
sample to zero to trim it (e.g. to display the first frame for a
slightly shorter period of time). However, we can't do this to audio
samples are they have an inherent duration and trimming them this
way is not possible.
#minor-release
PiperOrigin-RevId: 543420218
FallbackListener.onTransformationRequestFinalized() is called from the
AssetLoader thread for audio, and from the GL thread for video.
PiperOrigin-RevId: 542851284
The effects pipeline must receive images in the sRGB colorspace due to the color transfers applied in the shaders. Currently the burden to making sure images are in the right colorspaces falls onto apps. This CL ensures that this is not the case anymore.
PiperOrigin-RevId: 542323613
