I work on image segmentation and object detection research. In fact, my work can be found in the classifiers provided by the TensorFlow Object Detection package. I’ll be blunt, we suck at object detection and we really suck at object detection where there’re more than a handful of classes (there’s a reason we use top-k for evaluation) which is most problems. Furthermore, we’re entirely useless when it comes to problems that requires some interpretable taxonomy (e.g. make and model of a car). I encourage you to try some harder problems. :)