3D imaging – the Maxi-MARVIN

The Maxi-MARVIN is a high-throughput plant phenotyping device that generates three-dimensional plant models within milliseconds. The software was developed in-house by Wageningen University and Research. The MARVIN technology generates 3D models of plants and accurately determines the size and shape of leaves and stems plus the overall architecture in milliseconds. SMO will integrate the Maxi-MARVIN in the conveyor-based WIWAM phenotyping setup in the NPEC greenhouse.


Figure 1: schematic and picture of the Maxi MARVIN setup in the NPEC greenhouse.

The setup comprises 15 RGB cameras (Basler ace Classic, acA1920-25gc, 1920×1080 pixels, 8 mm lens, 16° and 30° horizontal and vertical field of view, respectively), arranged at three height levels with five cameras per level positioned approximately 1200 mm from the plant centre. Images are captured simultaneously with flash illumination to minimize exposure time and account for plant movement.

Figure 1. Schematic diagram of the Maxi-MARVIN setup at NPEC (left), Interior view of the Maxi-MARVIN with cameras highlighted in blue (right).

Comprehensive datasets

NPEC aims is to monitor individual plants over time to identify which genotypes perform optimally under specific environmental conditions. To achieve this, NPEC employs a comprehensive dataset strategy.

Comprehensive datasets integrate data from multiple plant sensors to maximize analytical output. In Greenhouse 1 & 2 of the NPEC facility, plants are sequentially transported through several sensor stations, generating datasets containing to several sensors. This result in a dataset with:

  • RGB and hyperspectral images (RGB sideview and SnapScan setup)
  • Voxel cubes clouds from the MAXI Marvin
  • Photosynthesis measurements (CropReporter setup)
  • Pot weight (Wiwam scale at watering station)
  • Climate data (Hoogendoorn Climate computer output)

A key advantage of the Maxi-Marvin is its ability to fuse multi-camera data through 3D reconstruction. Hyperspectral images and chlorophyll fluorescence data can be projected onto the 3D point clouds, enriching each point with spatial (XYZ) coordinates, spectral information and photosynthetic efficiency. On top of that the Maxi-Marvin has a point cloud post-processing pipeline. This pipeline derives plant traits through three sequential steps. (1) segmentation to classify leaves, stems, flowers and fruit structures; (2) instance segmentation to extract individual organs; and (3) skeletonization to determine the plant structure. The instance segmentation enables researcher to obtain spectral information for every plant organ. In addition, the plant structure allows automatic extraction of quantitative traits such as internode length, phyllotactic and leaf angle. An example output is schematic visualised in Figure 2.

Interested in the setup? Please have look at the published papers below. If you to know more please reach out to Rick van de Zedde via the Contact form on this website.
Figure 1 Example of deriving plant traits from point clouds. The giphy shows raw RGB point clouds, point cloud segmentation, instance segmentation of individual leaves, stem and lastly the plant structure.

MaxiMarvin software and specs:
Number of cameras 15
Camera Basler ace Classic, acA1920-25gc
Lens 8 mm
Resolution (w × h) 1080 × 1920
Horizontal field of view 16
Vertical field of view 30
3D point clouds technique Shape from silhouette (aka voxel carving)
Voxel dimensions (v1 – till March 2026) 400 × 400 × 700
Voxel dimensions (v2 – April 2026 onwards) 450 × 450 × 750
Voxel size 1 mm
Camera calibration → available TomatoWUR dataset 10.4121/e2c59841-4653-45de-a75e-4994b2766a2f
Basic notebook for data processing To be done

Related publications

Van Vlaardingen, M. et al. (2026).
Projecting 2D top-view of PSII efficiency onto 3D plant models to quantify PSII efficiency across canopy layers
Plant Methods.

Van de Zedde, R. (2026).
Unlocking the full potential plant
Prophyta – The Annual 2026, 22–23.

Smoleňová, K. et al. (2025).
Development of a tomato functional–structural plant model for digital twin applications
in silico Plants, 7.

Van Marrewijk, B. M. et al. (2025).
TomatoWUR: an annotated dataset of tomato plants to quantitatively evaluate segmentation, skeletonisation, and plant-trait extraction algorithms for 3D plant phenotyping
Data in Brief, 61, 111852.

Adebola, S. et al. (2025).
GrowSplat: Constructing Temporal Digital Twins of Plants with Gaussian Splats
2025 IEEE International Conference on Automation Science and Engineering, 1766–1773.

Van Marrewijk, B. M. et al. (2025).
3D plant segmentation: Comparing a 2D-to-3D segmentation method with state-of-the-art 3D segmentation algorithms
Biosystems Engineering, 254, 104147.

Xin, B. et al. (2024).
An automatic 3D tomato plant stemwork phenotyping pipeline at internode level based on tree quantitative structural modelling algorithm
Computers and Electronics in Agriculture, 227, 109607.

Xin, B. et al. (2023).
3D data-augmentation methods for semantic segmentation of tomato plant parts
Frontiers in Plant Science, 14, 1045545.

Golbach, F. et al. (2016).
Validation of plant part measurements using a 3D reconstruction method suitable for high-throughput seedling phenotyping
Machine Vision and Applications, 27, 663–680.

Koenderink, N. J. J. P. et al. (2009).
MARVIN: high speed 3D imaging for seedling classification
In Precision Agriculture ’09, 279–286.