Summary

This chapter discussed the pre-trained models available on the Detectron2 Model Zoo. Specifically, it specified the vital information to look for while selecting a pre-trained model for a computer vision task. It then provided the code snippets for developing an object detection application, an object instance segmentation application, a keypoint detection application, a panoptic segmentation application, and a semantic segmentation application. The code snippets for these applications are similar. Therefore, the code snippets are abstracted into typical methods for getting a model, performing inferences, and visualizing outputs. These methods can then be reused to develop different computer vision applications with Detectron2 quickly.

The cutting-edge models on Detectron2 are trained with many images and a vast amount of computation resources. Therefore, these models have high accuracies. Additionally, they are also trained on the datasets with class labels for everyday...