Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GrowSplat: Constructing Temporal Digital Twins of Plants with Gaussian Splats

Simeon Adebola, Shuangyu Xie, Chung Min Kim, Justin Kerr, Bart M. van Marrewijk, Mieke van Vlaardingen, Tim van Daalen, E. N. van Loo, Jose Luis Susa Rincon, Eugen Solowjow, Rick van de Zedde, Ken Goldberg

TL;DR

GrowSplat tackles the challenge of temporally reconstructing plant growth as a 4D digital twin from multi-view data. It uses Gaussian splats to represent plant geometry and a two-stage registration pipeline that aligns sequential scans through a coarse feature-based/fast global registration stage followed by ICP refinement. The approach is validated on NPEC greenhouse data for Sequoia and Quinoa, demonstrating detailed, time-consistent reconstructions that enable longitudinal trait analysis. This work advances high-fidelity, scalable digital twins for plant phenotyping, with potential impact on breeding and growth studies.

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/

GrowSplat: Constructing Temporal Digital Twins of Plants with Gaussian Splats

TL;DR

GrowSplat tackles the challenge of temporally reconstructing plant growth as a 4D digital twin from multi-view data. It uses Gaussian splats to represent plant geometry and a two-stage registration pipeline that aligns sequential scans through a coarse feature-based/fast global registration stage followed by ICP refinement. The approach is validated on NPEC greenhouse data for Sequoia and Quinoa, demonstrating detailed, time-consistent reconstructions that enable longitudinal trait analysis. This work advances high-fidelity, scalable digital twins for plant phenotyping, with potential impact on breeding and growth studies.

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/
Paper Structure (13 sections, 2 equations, 3 figures)

This paper contains 13 sections, 2 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. High-Throughput Industrial-Scale Plant Phenotyping
  4. 3D Reconstruction
  5. 4D Plant Modeling
  6. System Overview
  7. The Maxi-Marvin Setup
  8. Data Preprocessing for NerfStudio
  9. Problem Statement
  10. Assumptions
  11. Notation
  12. Transformation Model
  13. Method

Figures (3)

  • Figure 1: GrowSplat creates detailed 3D digital twins of plants with industrial-scale data and then constructs temporal digital twins over time.
  • Figure 2: Maxi-Marvin is an indoor system for plant phenotyping that constists of 15 calibrated static cameras. Plants are moved into the Maxi-Marvin using a conveyor belt and 15 images are taken. The system can be used for large plants up to a height of 70cm.
  • Figure 3: GrowSplat Digital Twins: Presented here are a side view for two different plants. Each column shows an RGB view of the 3D model.