mysterious

TODO

@@ -3,7 +3,9 @@
 - [x] camera switcher: discovered that it is broken https://github.com/aiortc/aiortc/issues/304 meanwhile Pion has been supporting it for 3 years https://github.com/pion/webrtc/pull/1527
 beginning of port from python/aiortc to go/pion ... maybe other issues around performance will be resolved too by doing so.
 - [x] write a go version of the compressed_vipc decoder demo
-- [ ] write a Pion-compatible vision IPC track by looking at https://github.com/pion/webrtc/blob/master/examples/rtp-to-webrtc/main.go and https://github.com/asticode/go-astiav/blob/master/examples/transcoding/main.go
+- [x] write a Pion-compatible vision IPC track by looking at https://github.com/pion/webrtc/blob/master/examples/rtp-to-webrtc/main.go and https://github.com/asticode/go-astiav/blob/master/examples/transcoding/main.go
+- [ ] debug why we dont see anything on the iOS side even though it appears RTP packets are flowing out
+- [ ] integrate bandwidth estimation https://github.com/pion/webrtc/blob/master/examples/bandwidth-estimation-from-disk/main.go
 - battery view
 - lowcam view
 - multicam

encoder.go

@@ -53,11 +53,14 @@ func NewEncoder(params EncoderParams) (ts *Encoder, err error) {
 
 	s.encCodecContext.SetHeight(params.Height)
 	s.encCodecContext.SetPixelFormat(params.PixelFormat)
-	s.encCodecContext.SetBitRate(1000)
+	s.encCodecContext.SetBitRate(200_000)
 	s.encCodecContext.SetSampleAspectRatio(params.AspectRatio)
 	s.encCodecContext.SetTimeBase(params.TimeBase)
 	s.encCodecContext.SetWidth(params.Width)
 
+	// Update flags
+	s.encCodecContext.SetFlags(s.encCodecContext.Flags().Add(astiav.CodecContextFlagGlobalHeader))
+
 	// Open codec context
 	if err = s.encCodecContext.Open(s.encCodec, nil); err != nil {
 		err = fmt.Errorf("encoder: opening codec context failed: %w", err)

main.go

@@ -21,12 +21,12 @@ func main() {
 		<-signal.PairWaitChannel
 	}
 	signal.OnPeerConnectionCreated = func(pc *webrtc.PeerConnection) {
-		// ReplaceTrack("road", pc)
+		ReplaceTrack("road", pc)
 	}
 
-	visionTrack, _ = NewVisionIpcTrack("roadEncodeData")
+	// visionTrack, _ = NewVisionIpcTrack("roadEncodeData")
-	videoTrack, _ := visionTrack.NewTrackRTP()
+	// _, _ = visionTrack.NewTrackRTP()
-	visionTrack.StartRTP(videoTrack)
+	// visionTrack.StartRTP()
 
 	for {
 		select {}
@@ -78,7 +78,7 @@ func ReplaceTrack(prefix string, peerConnection *webrtc.PeerConnection) {
 				log.Println(fmt.Errorf("main: peer connection closed due to error: %w", err))
 			}
 		} else if connectionState.String() == "connected" {
-			go visionTrack.StartRTP(videoTrack)
+			go visionTrack.StartRTP()
 		}
 	})
 }

vision_ipc_track.go

@@ -5,13 +5,12 @@ import (
 	"fmt"
 	"io"
 	"log"
-	"strings"
 	"time"
 
 	"github.com/asticode/go-astiav"
 	"github.com/commaai/cereal"
 	zmq "github.com/pebbe/zmq4"
-	"github.com/pion/rtp/v2"
+	"github.com/pion/rtp"
 	"github.com/pion/rtp/v2/codecs"
 	"github.com/pion/webrtc/v3"
 )
@@ -33,14 +32,15 @@ type VisionIpcTrack struct {
 	seenIframe bool
 	pkt        *astiav.Packet
 	f          *astiav.Frame
+	encoder    *Encoder
+	packetizer rtp.Packetizer
+	videoTrack *webrtc.TrackLocalStaticRTP
 
 	networkLatency float64
 	frameLatency   float64
 	processLatency float64
 	pcLatency      float64
 	timeQ          []int64
-
-	Frame chan *Frame
 }
 
 type Frame struct {
@@ -112,8 +112,6 @@ func NewVisionIpcTrack(name string) (track *VisionIpcTrack, err error) {
 		processLatency: 0.0,
 		pcLatency:      0.0,
 		timeQ:          []int64{},
-
-		Frame: make(chan *Frame),
 	}, nil
 }
 
@@ -146,17 +144,14 @@ func (v *VisionIpcTrack) Stop() {
 		return
 	}
 	Open = false
-	defer func() {
 	v.f.Free()
 	v.pkt.Free()
 	v.stream.decCodecContext.Free()
 	v.subscriber.Close()
 	v.context.Term()
-	}()
-	<-v.Frame
 }
 
-func (v *VisionIpcTrack) Start() {
+func (v *VisionIpcTrack) Start(OnFrame func(outFrame *Frame)) {
 	if Open {
 		return
 	}
@@ -166,6 +161,9 @@ func (v *VisionIpcTrack) Start() {
 		if len(msgs) > 0 {
 			for _, msg := range msgs {
 
+				v.pkt.Unref()
+				v.f.Unref()
+
 				evt, err := cereal.ReadRootEvent(msg)
 				if err != nil {
 					log.Fatal(fmt.Errorf("cereal read root event failed: %w", err))
@@ -272,33 +270,11 @@ func (v *VisionIpcTrack) Start() {
 					len(data),
 					v.name,
 				)
-				v.Frame <- &Frame{Frame: v.f, Roll: roll}
+				outFrame := &Frame{Frame: v.f, Roll: roll}
-
+				OnFrame(outFrame)
 			}
 		}
 	}
-	close(v.Frame)
-}
-
-func TestVisionIPCTrack(name string) {
-	track, err := NewVisionIpcTrack(name)
-	if err != nil {
-		log.Fatal(fmt.Errorf("main: creating track failed: %w", err))
-	}
-	defer track.Stop()
-
-	// Handle ffmpeg logs
-	astiav.SetLogLevel(astiav.LogLevelError)
-	astiav.SetLogCallback(func(l astiav.LogLevel, fmt, msg, parent string) {
-		log.Printf("ffmpeg log: %s (level: %d)\n", strings.TrimSpace(msg), l)
-	})
-
-	go track.Start()
-
-	for frame := range track.Frame {
-		// Do something with decoded frame
-		fmt.Println(frame.Roll)
-	}
 }
 
 func (v *VisionIpcTrack) NewTrackRTP() (videoTrack *webrtc.TrackLocalStaticRTP, err error) {
@@ -307,10 +283,11 @@ func (v *VisionIpcTrack) NewTrackRTP() (videoTrack *webrtc.TrackLocalStaticRTP,
 		MimeType:  webrtc.MimeTypeH264,
 		ClockRate: 90000,
 	}, "video", "pion")
+	v.videoTrack = videoTrack
 	return videoTrack, err
 }
 
-func (visionTrack *VisionIpcTrack) StartRTP(videoTrack *webrtc.TrackLocalStaticRTP) {
+func (visionTrack *VisionIpcTrack) StartRTP() {
 	encoderParams := EncoderParams{
 		Width:       visionTrack.Width(),
 		Height:      visionTrack.Height(),
@@ -327,23 +304,25 @@ func (visionTrack *VisionIpcTrack) StartRTP(videoTrack *webrtc.TrackLocalStaticR
 	}
 	defer encoder.Close()
 
-	go visionTrack.Start()
+	visionTrack.encoder = encoder
-	defer visionTrack.Stop()
 
 	// Create a H.264 payloader
 	payloader := &codecs.H264Payloader{}
 
 	// Create a RTP packetizer with some parameters
-	packetizer := rtp.NewPacketizer(
+	visionTrack.packetizer = rtp.NewPacketizer(
-		1600,
+		1200,
-		96,
+		0, // Value is handled when writing
-		1,
+		0, // Value is handled when writing
 		payloader,
 		rtp.NewRandomSequencer(),
+		visionTrack.videoTrack.Codec().ClockRate,
 	)
 
-	for visionTrack != nil {
+	visionTrack.Start(visionTrack.HandleFrameRTP)
-		for frame := range visionTrack.Frame {
+}
+
+func (v *VisionIpcTrack) HandleFrameRTP(frame *Frame) {
 	fmt.Println(frame.Roll)
 
 	// so right now frame is a raw decoded AVFrame
@@ -352,29 +331,22 @@ func (visionTrack *VisionIpcTrack) StartRTP(videoTrack *webrtc.TrackLocalStaticR
 
 	// we need to:
 	// 1. transcode to h264 with adaptive bitrate using astiav
-			outFrame, err := encoder.Encode(avframe)
+	outFrame, err := v.encoder.Encode(avframe)
 	if err != nil {
 		log.Println(fmt.Errorf("encode error: %w", err))
 
-				continue
+		return
 	}
 
 	// 2. create an RTP packet with h264 data inside using pion's rtp packetizer
-			log.Println(outFrame.Duration())
+	samples := uint32((1.0 / v.encoder.Stream.encCodecContext.TimeBase().ToDouble()) * float64(visionTrack.videoTrack.Codec().ClockRate))
-			rtpPackets := packetizer.Packetize(outFrame.Data(), uint32(outFrame.Duration()))
+	rtpPackets := v.packetizer.Packetize(outFrame.Data(), samples)
 
 	// 3. write the RTP packet to the videoTrack
 	// Loop over the RTP packets and send them to an output stream
-			for rtpPacketNo, rtpPacket := range rtpPackets {
+	for _, rtpPacket := range rtpPackets {
-				// Write the RTP packet to output
-				data, err := rtpPacket.Marshal()
-				if err != nil {
-					log.Println(fmt.Errorf("rtp marshall error: %w", err))
 
-					continue
+		err = v.videoTrack.WriteRTP(rtpPacket)
-				}
-
-				bytesWritten, err := videoTrack.Write(data)
 		if err != nil {
 			if errors.Is(err, io.ErrClosedPipe) {
 				// The peerConnection has been closed.
@@ -384,10 +356,7 @@ func (visionTrack *VisionIpcTrack) StartRTP(videoTrack *webrtc.TrackLocalStaticR
 			log.Println(fmt.Errorf("rtp write error: %w", err))
 		}
 
-				log.Printf("RTP(%d:%d)", rtpPacketNo, bytesWritten)
+		// log.Printf("RTP(%d)", rtpPacketNo)
 
 	}
-
-		}
-	}
 }