Setting up the TensorFlow Object Detection API

The TensorFlow Object Detection API is documented in detail at its official site https://github.com/tensorflow/models/tree/master/research/object_detection, and you should definitely check out its “Quick Start: Jupyter notebook for off-the-shelf inference” guide for a quick idea of how to use a good pre-trained model for detection in Python. But the documentation there is spread out over many different pages, making it sometimes difficult to follow. In this and the next sections, we’ll streamline the official documentation by reorganizing the important details documented in many different places and adding more examples and code explanations, and offer two step-by-step tutorials on:

  1. How to set up the API and use its pre-trained models for off-the-shelf inference
  2. How to retrain pre-trained models with the API for more specific detection tasks