VideoEncoder using inputSurfaceView to send to RTMPClient

[discovertalha]

Hello,

I have been trying to send a video feed directly from DeepAR but DeepAR allows only one mode at a time (renderToSurfaceView) or offScreenRendering which gives me processed Frames that dont function very well.

My primary goal is to directly feed the surface as that will probably work best for my case but the setInputSurface method that comes with the VideoEncoder doesnt work in my case. I am working with the 2.2.6 version of the RootEncoder. With using frame by frame approach I managed to somewhat convert the frames and send them through the VideoEncoder and the GetVideoData interface but the performance is abysmal and the frames get distorted because of DeepArs weird way of processing frames. Is there anyway we could directly just use the Surface that DeepAR is rendering its frames on ?

The current approach which I experimented around with is something like:

public void frameAvailable(Image frame) {
    if (frame == null) {
        Log.e(TAG, "frameAvailable: Received null frame");
        return;
    }

// renderer.renderImage(frame);
ByteBuffer buffer=frame.getPlanes()[0].getBuffer();
int[] byteArray=getArrayfromBytes(buffer);
if(byteArray.length>0){
ByteBuffer yuvBuffer=YUVBufferExtractor.convertImageToYUV(frame);
// byte[] yuvByteArr = YUVUtil.ARGBtoYUV420SemiPlanar(byteArray, frame.getWidth()+48, frame.getHeight());
byte[] yuvByteArr=yuvBuffer.array();
Frame yuvFrame =new Frame(yuvByteArr, 0, frame.getWidth()*frame.getHeight());
renderer.renderImageThroughFrame(yuvFrame, frame.getWidth(), frame.getHeight());
videoEncoder.inputYUVData(yuvFrame);
}
}

[pedroSG94]

Hello,

I did a demo implementation of the library working with deepAR here:
https://github.com/pedroSG94/demo-deepAR-streaming
You only need focus on this classes:
https://github.com/pedroSG94/demo-deepAR-streaming/tree/feature/fixing-streaming/app/src/main/java/com/example/streaming_deepar/streaming

[discovertalha]

It was working great, but now I'm facing an issue with the aspect ratio during streaming. It's not displaying correctly. Specifically, after the face is detected by DeepAR, the resulting stream seems to convert to a landscape orientation instead.

[pedroSG94]

Can you share a screenshot?

[discovertalha]

mobile side

[discovertalha]

[discovertalha]

when set
rtmpStream.getGlInterface().setAutoHandleOrientation(true);

stream side aspect ratio is working

[discovertalha]

        rtmpStream.getGlInterface().setAutoHandleOrientation(true);

mobile side

Image

[pedroSG94]

Ok, I see

You only have problems in the preview (mobile side).

Do you have a full code example with your modifications?

[discovertalha]

private boolean initializeDeepAR(String licenseKey, CameraResolutionPreset resolutionPreset) {
    try {
        int width = resolutionPreset.getWidth();
        int height = resolutionPreset.getHeight();
        deepAR = new DeepAR(activity);
        deepAR.setLicenseKey(licenseKey);
        deepAR.initialize(activity, this);
        deepAR.changeLiveMode(true);

        TextureRegistry.SurfaceTextureEntry entry = flutterPlugin.getTextureRegistry().createSurfaceTexture();
        tempSurfaceTexture = entry.surfaceTexture();
        tempSurfaceTexture.setDefaultBufferSize(width, height);
        surface = new Surface(tempSurfaceTexture);
        textureId = entry.id();
        var source = new DeepARSource(activity, licenseKey, deepAR, this);
        MicrophoneSource microphoneSource = new MicrophoneSource();
        rtmpStream = new RtmpStream(activity, this, source, microphoneSource);
        rtmpStream.getGlInterface().setAutoHandleOrientation(false);
        rtmpStream.getGlInterface().setIsPortrait(false); // Force portrait
        boolean prepared = rtmpStream.prepareVideo(
                width,
                height,
                1200 * 1000,
                30
        ) && rtmpStream.prepareAudio(
                44100,
                true,
                128 * 1000
        );

        if (prepared) {
            Log.d("RTMP", "Stream prepared successfully");
            rtmpStream.startPreview(surface, width, height);
            rtmpStream.getGlInterface().setPreviewResolution(width, height);
            // Add orientation check here
            rtmpStream.getGlInterface().setAutoHandleOrientation(true);
            boolean isAutoOrientation = rtmpStream.getGlInterface().getAutoHandleOrientation();
            Log.d("OrientationCheck", "Auto Orientation: " + isAutoOrientation);
            Log.d("OrientationCheck", "Current Orientation: " + rtmpStream.getGlInterface().getAutoHandleOrientation());
            rtmpStream.startStream("rtmp://ingest.global-contribute.live-video.net/app/live_176582760_BKfgmOZd8rlcKONDfGZK9MxXCsV6cD");
            Log.d("RTMP", "Streaming started successfully");
            return true;
        }
        return false;
    } catch (Exception e) {
        Log.e(TAG, "Error: " + e.getMessage(), e);
        return false;
    }
}
this was intitialized method

deep ar source file
package com.example.myapp
import ai.deepar.ar.AREventListener
import ai.deepar.ar.DeepAR
import android.content.Context
import android.graphics.SurfaceTexture
import android.view.Surface
import com.pedro.encoder.input.sources.video.VideoSource

/**
 * Created by pedro on 17/10/24.
 */
class DeepARSource(
  context: Context,
  licenseKey: String,
  deepARInstance:DeepAR,
  listener: AREventListener
): VideoSource() {
  private var running = false
  val deepAR=deepARInstance;
  override fun create(width: Int, height: Int, fps: Int, rotation: Int): Boolean {
    return true
  }

  override fun isRunning(): Boolean = running

  override fun release() {
    deepAR.release()
  }

  override fun start(surfaceTexture: SurfaceTexture) {
    deepAR.setRenderSurface(Surface(surfaceTexture), height, width)
    deepAR.startCapture()
    running = true
  }

  override fun stop() {
    deepAR.stopCapture()
    running = false
  }
}