Roboflow is an open-source dataset management platform, where users can import data, label it, and then have a model train on it. For my group's project, CS-FLARE, we used Roboflow to label brush, grass, trees, and leaf litter in our NESW Images as shown below.
When I initially began labeling images, I tried using bounding boxes for each picture like the leaf litter in the image above. This became a bit annoying as sometimes the bounding boxes would cut parts of objects like trees, but it was still faster than using the polygon tool to manually trace around objects. Eventually, I was introduced to the smart polygon tool which automatically created the polygons around objects and I could add or subtract to the initial polygon. At first, I thought this was a blessing as I could get the precision I wanted while still quickly labeling images. Unfortunately, because of some of my perfectionist tendencies, the smart polygon tool made labeling much more difficult for me. This was because sometimes the tool would select too little or too much of an object causing me to take time to have objects perfectly within the label. Take the image below as an example.
To me, what I have circled in red are areas that shouldn't be included in the labeling and annoy me a lot. To fix these areas though, it would maybe add on 1 or 2 minutes to my total labeling time as I would have to continuously add or subtract areas until the smart polygon tool understood how I wanted the shape of the label to look like. Even though that doesn't sound too bad, when you have 1000+ images to label, it adds up, especially when you could label an image like the one above in 20 seconds at the max. Because of small but noticeable areas that shouldn't be included in the labeling according to my standards, this process became incredibly tedious and time-consuming. Overall though, Roboflow is a great tool to use to annotate images and they even provide you with a model to train the images on if you aren't knowledgeable about AI. For these reasons, I recommend it to people who need to easily label their data, but personally, I would not want to use it again.
About the author, Ashvin is a rising senior from Frisco, Texas. This virtual internship is part of a collaboration between the Institute for Global Environmental Strategies (IGES) and the NASA Texas Space Grant Consortium (TSGC) to extend the TSGC Summer Enhancement in Earth Science (SEES) internship for U.S. high school (http://www.tsgc.utexas.edu/sees-internship/). This guest blog shares the NASA SEES Earth System Explorers virtual internship in 2024.
