GrowSplat: Constructing Temporal Digital Twins of Plants with Gaussian Splats
🤖 AI Summary
To address the challenge of temporally consistent 3D reconstruction in plant phenotyping—complicated by complex geometry, occlusions, and non-rigid growth—this paper proposes a growth-aware 4D digital twin modeling framework. Our method integrates 3D Gaussian Splatting with a two-stage registration pipeline (Fast Global Registration followed by ICP), enabling, for the first time, continuous spatiotemporally coherent reconstruction under non-rigid deformation. Given multi-view image sequences, it generates high-fidelity dynamic 3D models. Evaluated on the Netherlands Plant Eco-Phenotyping Center dataset, our approach successfully reconstructs fine-grained growth sequences of sequoia and quinoa. It achieves significant improvements in geometric completeness (+28.6%) and temporal consistency (41.3% reduction in temporal alignment error). This work establishes a scalable, high-accuracy digital twin foundation for automated, precision phenotyping and intelligent breeding.
Application Category
📝 Abstract
Accurate temporal reconstructions of plant growth are essential for plant phenotyping and breeding, yet remain challenging due to complex geometries, occlusions, and non-rigid deformations of plants. We present a novel framework for building temporal digital twins of plants by combining 3D Gaussian Splatting with a robust sample alignment pipeline. Our method begins by reconstructing Gaussian Splats from multi-view camera data, then leverages a two-stage registration approach: coarse alignment through feature-based matching and Fast Global Registration, followed by fine alignment with Iterative Closest Point. This pipeline yields a consistent 4D model of plant development in discrete time steps. We evaluate the approach on data from the Netherlands Plant Eco-phenotyping Center, demonstrating detailed temporal reconstructions of Sequoia and Quinoa species. Videos and Images can be seen at https://berkeleyautomation.github.io/GrowSplat/
Problem
Research questions and friction points this paper is trying to address.
Temporal reconstructions of plant growth for phenotyping
Handling complex geometries and non-rigid plant deformations
Building 4D digital twins using Gaussian Splatting
Innovation
Methods, ideas, or system contributions that make the work stand out.
Combines 3D Gaussian Splatting with alignment pipeline
Uses two-stage registration for precise plant modeling
Creates consistent 4D models of plant growth
🔎 Similar Papers
No similar papers found.
