Part 2: Developing Custom Object Detection Models
The second part is about getting your hands busy developing and fine-tuning custom detection models with data preparation, training, and fine-tuning steps. The data preparation step introduces common computer vision datasets and the code to download freely available images. Additionally, it discusses tools to label data, common annotation formats, and the code to convert from different formats to the one Detectron2 supports. It then goes into further detail about the architecture of a Detetron2 application using visualizations and code. After training a model, this part illustrates the steps to utilize TensorBoard to find insights about training before fine-tuning the trained models. For fine-tuning, this section provides a primer on deep-learning optimizers and steps to fine-tune Detectron2 solvers. For optimizing detection models specifically, this part includes the code to compute the suitable sizes and ratios parameters for generating...
