Custom Object Detection for GIS at UT-Libraries

My capstone project aims to see if there are machine learning solutions available for automating the creation of map annotations for The UT-Libraries Map Collection. Much of this project has been dedicated to appropriately preparing the image dataset for pre-existing ML object detection models in TensorFlow. This has involved creating Python scripts for prepping, sorting and converting the existing dataset into images and annotations that will work for an object detection ML environment. Some of the challenges have been related to resource limitations. These include disk space limits and Google Colab timing out before the model can finish its training.