AutoSplant: Segmenting 3D Plant Models

Department of Computer Science
University of Saskatchewan

Abstract

3DGS is a young technology, and although 3DGS itself is accessible with consumer cameras, processing techniques are still relatively inaccessible. Manual processing requires a lot of human labour, so robust open-source pipelines are needed to leverage this technology to its full potential for large scale tasks, such as plant phenotyping and crop analysis for plant breeders. We have developed an open-source 3DGS processing pipeline called AutoSplant that automates the processing steps for plant capture; AutoSplant takes a video or collection of images and outputs the plants within as isolated 3D Gaussian Splats, improving the accessibility of these techniques. We performed visual evaluations to compare the quality and evaluate the potential of such technology.

PGSR Training

Pipeline Images