This project explores how to reconstruct 3D pollen grain structures from only two orthogonal grayscale images, inspired by the image outputs of SwisensPoleno. We evaluate techniques such as Neural Radiance Fields (NeRF), Gaussian Splatting, and VAE-based 3D generation on synthetic and real-world datasets to determine their reconstruction capabilities in constrained scenarios.