Authors: Jordan Pierce

​

I spend a fair amount of time working with scientists and researchers who use point annotations for labeling benthic quadrat survey images. Points are useful for capturing useful statistics including community composition, without needing to label every individual pixel in each image (a very time-consuming process indeed).

​

While working with scientists at NOAA (Bryan Costa, Shay Viehman, Clint Edwards), I started to get requests for utilizing CoralNet in ways that the platform itself did not support. Having had experience with point annotations from my Master's, I started to build a number of tools all centered around this idea of point annotations for coral reef imagery. Since I was developing it from the ground up, it was easy to unify these tools into a cohesive "toolbox".

​

Originally the idea was to just build something like a "CoralNet-Crawler", which could be used to search for Sources containing useful data for one's own project. It would then download the *already publicly available data* from the website for the user. With this data (in addition to our own) we can do some other cool data science-y stuff with it ( interested students / interns please inquire).

 

Over time though we added other tools including those for sampling points on images, cropping patches, pre-training and fine-tuning image classification models locally (PyTorch). I also threw in the ability to train semantic segmentation models (of which the segmentation masks can be created automatically with existing points using SAM); because I want people to use my Master's work, I also included a basic script to run through the Metashape workflow, and also, using the segmentation masks, create classified versions of the 3D data products.

​

Hoping that these tools will be useful to other researchers outside of NOAA, we made it open-source on my GitHub. I'm very happy to hear others find it useful, and will gladly add additional features for those who request them. Cheers!

​

(Please cite and make pull requests 😎)