Tensorflow Object Detection API

Surfacing as a popular toolkit of machine learning technologies in early-mid 2017, the Tensorflow object detection API, released by Google, is an open source framework for object detection related tasks used for training both Single Shot Detector (SSD) and regional-Convolutional Neural Network (R-CNN) models from their model zoo (Mustamo 2018). The Tensorflow API was essentially purposed to offer scalability and potential for device deployment by Google. More importantly, Google prepared Tensorflow tools with necessary support for leading methods such as Multibox/SSD, and Fast/Faster R-CNN, which will be discussed further in this paper. The object detection API was created with an order/hierarchy of levels ranging from deployment to simple box operations. A low level API generally consists of box operations, Box representations, Target Assignment, and Visualization operations. A high level API is comprised of the heart-core structure of meta-architectures including SSD, Faster R-CNN, etc. Eventually, serving and deployment on technologies such as Jupyter Notebook and android is reached once training and visualization has been finalized. Figure 1.1 reveals an example output from the Jupyter Notebook object detection demo application.