ECCV 2026 · 4D Plant Growth

GrowFields

Compositional 4D Neural Fields for Topology-Changing Plant Growth

1 ETH Zürich 2 Swiss Data Science Center 3 Martin-Luther-Universität Halle-Wittenberg
Accepted at ECCV 2026
🌱 TL;DR

GrowFields models 4D plant growth by decomposing a plant into organs and evolving them with a shared, latent-conditioned neural velocity field that learns cross-organ growth priors while handling changing topology.

Overview of GrowFields: segmented point clouds are decomposed into organs, aligned to canonical frames, evolved by a shared latent-conditioned neural velocity field, and recomposed into a dense 4D plant reconstruction.
Overview of GrowFields. Given a time series of segmented plant point clouds, the plant is decomposed into organs, each aligned to a canonical coordinate frame (A). A single shared neural velocity field fθ, conditioned on a learnable per-organ latent code zi, integrates each organ's trajectory over time (B). Organ motions are recomposed in the global frame to produce a temporally coherent, dense 4D reconstruction (C), from which continuous organ-level growth traits can be read off (D).

Quantifying plant growth dynamics from sparse longitudinal 3D observations is fundamental for agriculture and plant sciences. Yet, plants pose unique challenges: they undergo intricate non-rigid deformations, exhibit changing topology as new organs emerge, and often lack explicit temporal correspondences between consecutive acquisitions due to newly formed tissue. Methods designed for general scenes struggle to model the topology changes and asynchronous organ growth characteristic of plants. To address this, we introduce GrowFields, a compositional dynamic neural field for organ-aware 4D plant growth modelling from point cloud time series. Our approach decomposes a plant into its constituent organs and aligns each into its own canonical coordinate frame, isolating intrinsic growth patterns from global plant motion. We then learn a shared continuous neural deformation field that models temporal dynamics across all organs, conditioned on learnable per-organ latent codes capturing organ identity and growth characteristics. The resulting modular yet unified representation naturally accommodates the asynchronous development of plant organs while remaining grounded in the practical setting of organ-level plant tracking. We evaluate GrowFields on growth sequences from four plant species, assessing geometric fitting and organ tracking with manually annotated leaf-tip trajectories, and demonstrate consistent improvements in spatial precision, temporal coherence, and morphological fidelity over a range of existing representations.


A modular dynamical system for plant growth

Plant growth is inherently modular — each organ evolves by its own geometry, yet organs grow in coordination as one organism. GrowFields mirrors both properties: it canonicalises organs to isolate intrinsic growth, then models their dynamics with a single shared field conditioned on per-organ latent codes, before recomposing the full plant in the global frame.

1

Organ canonicalisation

Each organ is decomposed from the plant and aligned to a canonical frame via PCA, removing global pose so the model sees only intrinsic shape and growth.

2

Shared latent-conditioned field

A single SIREN velocity field, conditioned on per-organ latent codes, is integrated as a Neural ODE — sharing priors across organs while specialising to each trajectory.

3

Topology-changing composition

Organs are reprojected and merged in the global frame. New organs join the shared system upon first observation, yielding coherent, topology-changing 4D plants.

4

Leaf-tip benchmark

We contribute an open evaluation protocol with newly annotated leaf-tip trajectories across four species, measuring both geometric fit and organ-level tracking.


State-of-the-art reconstruction & tracking

Across eight growth sequences from four species, GrowFields achieves the best full-plant geometric fit (Chamfer distance) and the most accurate leaf-tip tracking (end-point error), outperforming global deformation fields, 4D SDFs, point-based, and part-aware baselines.

Method CD (mm²) ↓ EPE (mm) ↓
Full-plant baselines
NDF2229.242.9
NVFi633.543.3
DSR134.5
CanFields615.744.4
DPF2.4715.15
Part-aware methods
COAP1.694.23
Ours (per-organ MLPs)0.932.37
Ours (full)0.861.47

Full-plant reconstruction, mean over eight sequences (5 runs each). Full-plant baselines take no segmentation input; part-aware methods use the same ground-truth organ segmentation as ours. DSR is a 4D-implicit method with no deformation field, so EPE is undefined.

Qualitative full-plant reconstruction: ground truth, GrowFields (full), and per-organ MLP variant across days 1 to 7. Qualitative full-plant reconstruction baselines: CanFields, DPF, and COAP across days 1 to 7.
Qualitative full-plant reconstruction. Colours denote organ identities from the ground-truth segmentation (except CanFields, which uses its own internal motion segmentation). GrowFields' shared latent-conditioned field resolves organ boundaries correctly and follows growth faithfully, where global and per-organ baselines drift or blur part identities.

4D dense interpolation

Because growth is modelled as a continuous velocity field, GrowFields can be sampled at any point in time — densifying a handful of daily scans into a smooth, temporally coherent trajectory of organ development.

Dense temporal interpolation of a maize plant by GrowFields: a smooth fan of many interpolated frames rolled out continuously from the first observation.
Dense interpolation. A continuous rollout of maize_control_plant2 from the first observation (left) forward through time (right). The markers below trace the time axis: each observed frame (large green dot) is followed by seven densely interpolated frames (small grey dots) predicted by the shared field. The whole trajectory is reconstructed as one smooth, temporally coherent sequence; colours denote organ identities.

Temporal extrapolation

Rolled out beyond the training horizon, GrowFields predicts plausible continued organ elongation and outperforms the part-aware baseline on every sequence — while geometric accuracy naturally degrades without generative priors for unseen morphology.

Temporal extrapolation of a maize plant: GrowFields trained on all but the last three timesteps and rolled out across the full sequence, with the held-out frames marked by a dashed box.
Qualitative extrapolation — maize. GrowFields on maize_control_plant2, trained on all but the last three timesteps and rolled out from the first observation across the full sequence (left → right). The extrapolation region (dashed box) marks the three held-out frames; the model anticipates continued organ elongation, though geometric accuracy degrades without generative priors. Colours denote organ identities.
Temporal extrapolation of a tomato plant: a long dense rollout by GrowFields with the held-out frames marked by a dashed box.
Qualitative extrapolation — tomato. A longer-horizon rollout on tomato1_heat_plant3. GrowFields tracks the coordinated growth of stem and leaves across many timesteps and continues plausibly into the extrapolation region (dashed box, held-out frames), including the newly emerging organ at the apex.

Robustness to real-world inputs

🌱

No ground-truth labels needed. Re-running GrowFields on automatically segmented point clouds (PSegNet + TrackPlant3D) yields a mean CD 0.86 / EPE 1.42 — matching the 0.86 / 1.47 obtained with ground-truth segmentation, and confirming the method holds under realistic preprocessing.

GrowFields reconstruction of a maize plant using automatically segmented organ labels as input, recovering coherent organ-level growth despite noisier labels.
Automated segmentation. GrowFields on maize_control_plant2 driven by automatically segmented organ labels. Despite noisier labels than the ground-truth segmentation, the method still recovers coherent organ-level growth; colours denote the input organ identities.

BibTeX

@inproceedings{gajardo2026growfields, title = {Grow{F}ields: {C}ompositional {4D} {N}eural {F}ields for {T}opology-{C}hanging {P}lant {G}rowth}, author = {Gajardo, Joaquin and Volpi, Michele and Mihajlovic, Marko and Tang, Siyu and Roth, Lukas and Prokudin, Sergey}, booktitle = {European Conference on Computer Vision (ECCV)}, year = {2026} }