AI-Generated 3D leaf models advance precision plant phenotyping

By creating synthetic “leaf point clouds,” the method dramatically reduces the need for manual measurements and enables more accurate, scalable trait estimation in plant phenotyping.

In recent years, 3D plant phenotyping has emerged as a promising field for understanding crop structure and productivity. However, accurately estimating leaf traits remains challenging because obtaining ground-truth data requires time-consuming manual work by experts. Traditional image-based methods capture only 2D features and struggle to represent leaf curvature and geometry, while 3D approaches are constrained by limited labeled data for training. As a result, most algorithms either rely on rule-based models or generate synthetic data that lack real-world realism. These challenges highlight the need for a scalable and automated solution to produce high-quality, labeled 3D data for plant trait estimation.

A study (DOI: 10.1016/j.plaphe.2025.100071) published in Plant Phenomics on 16 June 2025 by Gianmarco Roggiolani ’s team, University of Bonn, introduces a generative model capable of producing lifelike 3D leaf point clouds with known geometric traits, accelerating crop improvement and optimize yield predictions through data-driven modeling.

The research team trained a 3D convolutional neural network to learn how to generate realistic leaf structures from skeletonized representations of real leaves. Using datasets from sugar beet, maize, and tomato plants, they extracted the “skeleton” of each leaf—the petiole and main and lateral axes that define its shape—and then expanded these skeletons into dense point clouds using a Gaussian mixture model. The neural network, designed as a 3D U-Net architecture, predicts per-point offsets to reconstruct the complete leaf shape while maintaining its structural traits. A combination of reconstruction and distribution-based loss functions ensures that the generated leaves match the geometric and statistical properties of real-world data. To validate the method, the researchers compared their synthetic dataset against existing generative approaches and real agricultural data using metrics such as the Fréchet Inception Distance (FID), CLIP Maximum Mean Discrepancy (CMMD), and precision–recall F-scores. The generated leaves showed high similarity to real ones, outperforming alternative datasets produced by agricultural simulation software or diffusion models. Importantly, when the synthetic data were used to fine-tune existing leaf trait estimation algorithms, such as polynomial fitting and principal component analysis-based models, the accuracy and precision of trait prediction improved substantially. Tests conducted on the BonnBeetClouds3D and Pheno4D datasets confirmed that models trained with the new synthetic data estimated real leaf length and width more accurately and with lower error variance. The researchers also demonstrated that their approach could generate diverse leaf shapes conditioned on user-defined traits, allowing for robust benchmarking and model development without costly manual labeling.

This study represents a significant step toward automating 3D plant phenotyping and reducing the bottleneck caused by limited labeled data. By enabling realistic data generation based on real plant structures, the method provides a foundation for building, testing, and improving trait estimation algorithms in agriculture. Future work will expand this approach to handle more complex morphologies, such as compound leaves, and integrate it with plant growth models to simulate phenotypic changes across development stages. The team also envisions the creation of open-access libraries of synthetic yet biologically accurate plant datasets to support research in sustainable agriculture, robotic phenotyping, and crop improvement under climate challenges.

###

References

DOI

10.1016/j.plaphe.2025.100071

Original URL

https://doi.org/10.1016/j.plaphe.2025.100071

Funding information

This work has partially been funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy, EXC-2070 – 390732324 – PhenoRob.

About Plant Phenomics

Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.