Can you train the Tensorflow Object Detection API to detect parts of an object?

Question

I have set up the Tensorflow Object Detection API and I want to start training my own models. I have my training images ready and I am ready to start creating label XML's.

I want to train the model to recognise lots of different parts of a bicycle. So wheels, wheel axles, seat, handlebars, individual pedals, etc. Is that feasible? How granular can you go in deciding what an object is for the detector?

Answer 1

You can achieve this. Your problem statement is like any object detection model which is capable of detecting individual objects like the COCO model which can detect up to 90 objects like cat, dog, car, bicycle etc., Although in your case these are parts of the of same bigger object(bicycle), its possible using same approach of object detection in tensorflow.