A simple Swift package for video encoding and decoding with Annex-B adaptors optimized for transfering video over a network.
Example data flow when using
Transcoding
Transcoding
is used for video encoding and decoding in Castaway, an app that streams HDMI capture devices from an iPad or Mac to a nearby Vision Pro.
VideoEncoder
is an object that takes CMSampleBuffer
s containing CVPixelBuffer
s and outputs a stream of CMSampleBuffer
s containing CMBlockBuffer
s containing compressed H264/HEVC data. VideoEncoder
is initialized with a Config
, with presets for live capture, active transcoding, background transcoding, and ultra low latency following Apple's recommendations.
let videoEncoder = VideoEncoder(config: .ultraLowLatency)
encoderStreamTask = Task {
for await encodedSampleBuffer in videoEncoder.encodedSampleBuffers {
// encodedSampleBuffer: CMSampleBuffer > CMBlockBuffer
}
}
videoEncoder.encode(sampleBuffer)
VideoEncoder
is an object that takes CMSampleBuffer
s containing CMBlockBuffers
s containing compressed H264/HEVC data and outputs a stream of CMSampleBuffer
s containing CVPixelBuffer
s. VideoDecoder
is initialized with a Config
containing various optional decompression settings.
let videoDecoder = VideoDecoder(config: .init(realTime: true))
decoderStreamTask = Task {
for await decodedSampleBuffer in videoDecoder.decodedSampleBuffers {
// decodedSampleBuffer: CMSampleBuffer > CVPixelBuffer
}
}
videoDecoder.decode(sampleBuffer)
VideoEncoderAnnexBAdaptor
and VideoDecoderAnnexBAdaptor
can be used to convert compressed CMSampleBuffer
s to and from an Annex B (ITU-T-REC-H.265) byte stream. This is ideal for sending compressed video data over a network.
In this example, video frames from a capture device are encoded and decoded as a Annex-B data stream optimized for low latency.
let videoEncoder = VideoEncoder(config: .ultraLowLatency)
let videoEncoderAnnexBAdaptor = VideoEncoderAnnexBAdaptor(
videoEncoder: videoEncoder
)
let videoDecoder = VideoDecoder(config: .init(realTime: true))
let videoDecoderAnnexBAdaptor = VideoDecoderAnnexBAdaptor(
videoDecoder: videoDecoder,
codec: .hevc
)
videoEncoderTask = Task {
for await data in videoEncoderAnnexBAdaptor.annexBData {
// send data over network or whatever
}
}
videoDecoderTask = Task {
for await decodedSampleBuffer in videoDecoder.decodedSampleBuffers {
// here you have a received decoded sample buffer with image buffer
}
}
receivedMessageTask = Task {
// Replace `realtimeStreaming.receivedMessages` with however you receive encoded data packets
for await (data, _) in realtimeStreaming.receivedMessages {
videoDecoderAnnexBAdaptor.decode(data)
}
}
captureSessionTask = Task {
// Replace `captureSession.pixelBuffers` with your video data source
for await pixelBuffer in captureSession.pixelBuffers {
videoEncoder.encode(pixelBuffer)
}
}
- Currently
VideoDecoderAnnexBAdaptor
only supports decoding full NALU's, meaning if you are passing data over the network or some stream you must ensure you receive full video frame packets, not just an arbitrarily sized data stream. When using Network.framework, for example, you would use a customNWProtocolFramerImplementation
to receive individual messages. - There are a number of instances where either the encoder or decoder needs to be reset during an application lifecycle. The encoder and decoder automatically handle resetting after an iOS app has been backgrounded, but you may need to handle other cases by calling
encoder/decoder.invalidate()
. For example, if you maintain one encoder and disconnect then reconnect from a peer/decoder, the encoder will need to be invalidated to ensure it sends over the H264/HEVC parameter sets again. Otherwise the decoder will not be able to decode frames, as the encoder is optimized to only send SPS/PPS/VPS when necessary.