Minimal labels, maximal tracking: new AI framework follows every plant organ in 3D

Understanding how each individual leaf, stem, and bud grows and changes over time is essential for breeding better crops, but traditional tracking methods are either extremely laborintensive or cannot operate in real time. Researchers have now developed 3DOGT (3D Organ Growth Tracking), a new artificial intelligence framework that automatically tracks every plant organ throughout an entire growth sequence using only the organ labels from the very first moment. This approach achieves an average tracking accuracy of 88.10% across multiple crop species, outperforming existing methods while requiring a fraction of the manual annotation effort.

Modern plant phenotyping increasingly relies on timeseries 3D point cloud data—essentially a series of snapshots capturing the plant's shape at each growth stage. However, tracking individual organs across these snapshots remains a major bottleneck. Most current methods follow a "tracking by segmentation" strategy, which depends on having complete, accurate organ labels for every single time point—a process that demands weeks of painstaking human annotation. Other approaches use "backtracing" mechanisms, which require collecting the entire growth sequence before any analysis can begin, making realtime monitoring impossible. Due to these challenges, there is an urgent need for a tracking approach that minimizes reliance on manual segmentation while still enabling realtime, continuous monitoring.

A team of researchers from the School of Information and Intelligent Science at Donghua University in Shanghai has published (DOI: 10.1016/j.plaphe.2026.100181) a study on this framework in the journal Plant Phenomics on September 10, 2026. The 3DOGT (3D Organ Growth Tracking) framework introduces a fundamentally new paradigm, performing organ tracking and segmentation simultaneously with only the initial moment's organ labels as a starting point.

The 3DOGT framework works by iteratively processing pairs of adjacent point clouds in chronological order, propagating organ labels forward through the entire growth sequence. It integrates several cuttingedge techniques to achieve this. A pretrained deeplearning module called 3DNOD (3D New Organ Detection) identifies emerging buds and new leaves, assigning them fresh, independent labels to prevent them from being confused with older organs. For mature organs, a sophisticated labelbroadcasting system is employed. First, bidirectional point cloud registration aligns the adjacent moments both forward and backward in time using Coherent Point Drift (CPD). The aligned point clouds are then fed into a semisupervised Graph Convolutional Network (GCN), which propagates the sparse labels from the previous time point to the unlabeled points of the new one. A final refinement step corrects any residual label errors. The framework also incorporates an inhomogeneous downsampling strategy that preserves more points for smaller organs, reducing tracking errors where they are most likely to occur. Once merged with the newly detected organs, the complete, labeled point cloud for the next moment is produced, allowing the process to continue uninterrupted.

The authors highlighted the transformative potential of their approach: "We wanted to push the limits of how little information a tracking system actually needs to work reliably. With 3DOGT, we found that a single labeled snapshot is enough to follow an entire plant's growth journey. Watching the system automatically pick up a tiny new bud that even our human annotator had missed was a thrilling moment—it really proved that the framework sees what we sometimes overlook. For plant scientists, this means they can now monitor growth dynamics at scale without spending months on manual labeling, and for breeders, it opens the door to much faster, datadriven decisionmaking."

The ability to track every organ's growth automatically and in real time has farreaching implications for agriculture and fundamental plant biology. For highthroughput phenotyping, 3DOGT can dramatically accelerate breeding programs by enabling rapid, quantitative assessment of how different genotypes respond to environmental conditions such as drought, heat, or disease pressure. Its robustness against noisy sensor data and missing parts makes it wellsuited for deployment in realworld field conditions. Moreover, the framework's capacity to detect not only new organs but also sudden organ loss provides a powerful new tool for studying plant stress responses and senescence patterns, potentially guiding the development of more resilient crop varieties in an era of climate uncertainty.

###

References

DOI

10.1016/j.plaphe.2026.100181

Original Source URL

https://doi.org/10.1016/j.plaphe.2026.100181

Funding information

This work was supported by Donghua University 2025 Cultivation Project of Discipline Innovation (xkcx-202505).

About Plant Phenomics

Plant Phenomics publishes breakthrough research that integrates genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics and computer science to advance plant phenotyping—from the cellular level to entire populations. The journal tackles key scientific challenges in phenomics by showcasing cuttingedge technologies for data acquisition, management, interpretation and modelling. Its scope spans practical applications across all plant sciences, aiming to translate novel phenotyping tools into realworld solutions for breeding, crop management and fundamental plant biology. By fostering interdisciplinary innovation, Plant Phenomics serves as a hub for researchers developing highthroughput, accurate and scalable methods to decipher complex plant traits, ultimately accelerating progress in agriculture and plant science.