Abstract
The ability to automatically build 3D digital twins of plants from images hascountless applications in agriculture, environmental science, robotics, andother fields. However, current 3D reconstruction methods fail to recovercomplete shapes of plants due to heavy occlusion and complex geometries. Inthis work, we present a novel method for 3D reconstruction of agriculturalcrops based on optimizing a parametric model of plant morphology via inverseprocedural modeling. Our method first estimates depth maps by fitting a neuralradiance field and then employs Bayesian optimization to estimate plantmorphological parameters that result in consistent depth renderings. Theresulting 3D model is complete and biologically plausible. We validate ourmethod on a dataset of real images of agricultural fields, and demonstrate thatthe reconstructions can be used for a variety of monitoring and simulationapplications.