I use the AVCaptureSession to capture the video and uses AVCaptureVideoDataOutput / AVCaptureAudioDataOutput to collect the samples upon getting the samples, I parse them to the AVAssetWriter and to a CoreML YOLO class, that makes a prediction of the images. When I do this, it looks like the CoreML uses so many resources that the frames is dropped so the video file becomes unusable. I have tried with all kind of multi-thread but haven't been successful yet. Help Please In another settings I do the same for Speech Recognition, and it works like a charm