What this book covers
Chapter 1, An Introduction to Detectron2 and Computer Vision Tasks, introduces Detectron2, its architectures, and the computer vision tasks that Detectron2 can perform. Additionally, this chapter provides the steps to set up environments for developing computer vision applications using Detectron2.
Chapter 2, Developing Computer Vision Applications Using Existing Detectron2 Models, guides you through the steps to develop applications for computer vision tasks using state-of-the-art models in the Detectron2 Model Zoo. Thus, you can quickly develop practical computer vision applications without having to train custom models.
Chapter 3, Data Preparation for Object Detection Applications, discusses the steps to prepare data for training models using Detectron2. Additionally, this chapter covers the techniques to convert standard annotation formats to the data format required by Detectron2 in case the existing datasets come in different formats.
Chapter 4, The Architecture of the Object Detection Model in Detectron2, dives deep into the architecture of Detectron2 for the object detection task. This chapter is essential for understanding common terminologies when designing deep neural networks for vision systems.
Chapter 5, Training Custom Object Detection Models, provides steps to prepare data, train an object detection model, select the best model, and perform inferencing object detection tasks. Additionally, it details the development process of a custom trainer by extending the default trainer and incorporating a hook into the training process.
Chapter 6, Inspecting Training Results and Fine-Tuning Detectron2's Solver, covers the steps to use TensorBoard to inspect training histories. It utilizes the codes and visualizations approach for explaining the concepts behind Detectron2’s solvers and their hyperparameters. The related concepts include gradient descent, Stochastic gradient descent, momentum, and variable learning rate optimizers.
Chapter 7, Fine-Tuning Object Detection Models, explains how Detectron2 processes its inputs and provides codes to analyze the ground-truth boxes from a training dataset and find appropriate values for the anchor sizes and ratio configuration parameters. Additionally, this chapter provides the code to calculate the input image pixels' means and standard deviations from the training dataset in a rolling manner. The rolling calculations of these hyperparameters are essential if the training dataset is large and does not fit in the memory.
Chapter 8, Image Data Augmentation Techniques, introduces Detectron2’s image augmentation system with three main components: Transformation, Augmentation, and AugInput. It describes classes in these components and how they work together to perform image augmentation while training Detectron2 models.
Chapter 9, Applying Train-Time and Test-Time Image Augmentations, introduces the steps to apply these existing classes to training. This chapter also explains how to modify existing codes to implement custom techniques that need to load data from different inputs. Additionally, this chapter details the steps for applying image augmentations during test time to improve accuracy.
Chapter 10, Training Instance Segmentation Models, covers the steps to construct a dataset in the format supported by Detectron2 and train a model for a segmentation task. This chapter also utilizes the codes and visualizations approach to explain the architecture of an object segmentation application developed using Detectron2.
Chapter 11, Fine-Tuning Instance Segmentation Models, introduces PointRend, a project inside Detectron2 that helps improve the sharpness of the object’s boundaries. This chapter also covers the steps to use existing PointRend models and to train custom models using PointRend.
Chapter 12, Deploying Detectron2 Models into Server Environments, walks you through the steps in an export process to convert Detectron2 models into deployable artifacts. This chapter then provides the steps to deploy the exported models into the server environments.
Chapter 13, Deploying Detectron2 Models into Browsers and Mobile Environments, introduces the ONNX framework. It is extremely helpful when deploying Detectron2 models into browsers or mobile environments is needed. This chapter also describes D2Go for training, quantizing lightweight models extremely useful for deploying into mobile or edge devices.
