Semantic segmentation can be really ideal if you don’t have copious amounts of source data. While more well-labeled data is always a good thing, if you have a limited amount for your project, you can get more actionable information for your models from every single image.
Unlike bounding boxes which can capture a lot of white space and additional noise, leading to confusion in vision models, polygons are far more precise. You’ll often see them used for everything from aerial imagery to medical research.
Semantic segmentation can be really ideal if you don’t have copious amounts of source data. While more well-labeled data is always a good thing, if you have a limited amount for your project, you can get more actionable information for your models from every single image.
Labeling every part of an image so that every pixel is accounted for, the output here encodes ontology categories into an R,G,B pixel