This post is going to demonstrate how to do data augmentation for computer vision using the albumentations library. The exact data augmentations you use are going to be specific to your use-case. For example, if you’re training on overhead imagery the augmentations you use will probably be somewhat different than on an ImageNet-like dataset (although there will also be considerable overlap).

This post is a quick walkthrough of the different data augmentation methods available in Detectron2 and their utility for augmenting overhead imagery. I’ll also go over a quick way to implement them.

This post contains details of how I set up my shell and environment. I use Windows, Mac, and Linux on a daily basis, so I have different setups for different purposes, but I try to make them similar when I can.

This post contains details of how I set up my shell and environment. I use Windows, Mac, and Linux on a daily basis, so I have different setups for different purposes, but I try to make them similar when I can. You can see the software I use and how I customize it in the linked posts; this post will focus on setup.

This post contains details of how I set up my shell and environment. I use Windows, Mac, and Linux on a daily basis, so I have different setups for different purposes, but I try to make them similar when I can. You can see the softwhere I use and how I customize it in the linked posts; this post will focus on setup.

I ran into an interesting issue when debugging in Python that I thought was worth sharing. It came up when I was visualizing some results from an object detector. I had a class for an object detector and one of the things it would check was that each prediction was associated with a valid object in the object dictionary. Here is a mock-up of the relevant parts (this is not the real class, just a toy example so as not to distract from the point).

This posts contains some of my notes from switching to PyTorch after having worked with TensorFlow and Keras for a long time.