AI-generated images improved the performance of a standard image-recognition model when real training photographs were limited, according to a peer-reviewed computational modeling study by researchers at North Carolina State University.
The experiments focused on 20 common North American tree species and combined synthetic images with photographs from iNaturalist and the Auto Arborist Dataset.
The researchers trained a standard image-recognition model using different combinations of real and AI-generated tree images, then assessed its ability to identify the species represented in the training data.
Real photographs also capture variation that generated images may miss, including leaf shape, bark texture and growth form.
Subject of Research: Computational testing of synthetic tree imagery for training species-recognition models with limited real-image data Article Title: Study suggests AI-generated images could support conservation, but real-world data remains essential Article References: Study suggests AI-generated images could support conservation, but real-world data remains essential.