Biodiversity conservation efforts increasingly rely on large-scale image repositories for species classification and monitoring. This project addresses issues of class imbalance in such repositories to support more effective classification. My role involves balancing datasets through synthetic data generation and determining optimal strategies for integrating generative AI in biodiversity research. I’ll experiment with different diffusion model architectures and dataset parameters in generation, then quantify outcomes through model accuracy and generalization on out-of-distribution data. The goal is to introduce a validated biodiversity dataset, evaluate the potential of GenAI in ecological contexts, and propose best practices for data augmentation.

My goal through participating in this SuperUROP is to actively contribute to biodiversity conservation efforts using state of the art generation models. Being able to take more ownership of a project will allow me to grow my passion for research and practice both technical skills in computer vision and soft skills in communication.