As we mentioned in the first section, YOLO2 (https://pjreddie.com/darknet/yolo) is another cool object-detection model using a different approach from the RCNN family. It uses a single neural network to divide an input image into fixed-size regions (but no region proposals as in the RCNN-family methods) and predict bounding boxes, classes, and probabilities for each region.
The TensorFlow Android example app has sample code for using a pre-trained YOLO model, but there’s no iOS example. As YOLO2 is one of the fastest object-detection models and also pretty accurate (see the mAP comparison of it with SSD models at its website), it’s worth taking a look at what it takes to use it in an iOS app.
YOLO uses a unique open source neural network framework called Darknet (https://pjreddie.com/darknet) to train its models....
