- make
VNCoreMLRequest
every new videoframe (AVCaptureVideoDataOutputSampleBufferDelegate
) - use
VNSequenceRequestHandler
- using
MobileNet.mlmodel
(https://arxiv.org/abs/1704.04861v1) for object classification
- figure out the best way to recognize multiple objects ( a better model like https://pjreddie.com/darknet/yolo/ , or process frame as a grid of small images )
- bounding boxes
- object tracking