When training a custom neural net having large data sets improves accuracy. Often these data sets are open source, but otherwise building your own requires a custom tool. The tool we have created is an image annotator for establishing the ground truth. My task was to make the annotator as user friendly as possible and reduce the cost of entry for other data scientists to use. Improvements to the user interface where small, autofocusing text boxes, keyboard shortcuts to submit HTML forms, and system status indicators for showing the current state of the website. One of the biggest performance improvements was adding a bounding box to the annotator. The bounding box would allow a user to constrain the image to just the relevant pixels for the desired annotation. Meaning that the computational workload of the image segmentation algorithm was reduced and processing times where decreased. Machine learning image segmentation takes a lot of computational power so having a machine hooked up to a graphics card is necessary to achieve reasonable performance and speed. To lower the cost of entry, the website is hosted with CyVerse. Users can be given their own machine instance with a GPU to run the annotator for free.