Project Page | Mesh Dataset

Demeter is a plant parametric models that is learned from 3D scans of real-world plants. It explicitly models the plant as a graph of stem and leaf.

The processed 3d parametric plant samples are already included in the code.

The raw soybean mesh data can be found in this google drive link. It contains 607 unprocessed meshes, which can be used for 3D generation/representation learning. The main stem are aligned to y-axis and the bottom tip lies in (0,0,0). We will release the correspondent 2D images soon.

• Linux
• Python 3.11
• CUDA 12.1
• Pytorch 2.5.0

Install PyTorch and other dependencies.

conda create -n demeter python=3.11 -y
conda activate demeter
pip install torch==2.5.0 torchvision==0.20.0 torchaudio==2.5.0 --index-url https://download.pytorch.org/whl/cu121

# basic dependencies for decoding
pip install -r requirements.txt

for reconstruction from 3d point cloud, it is recommended to create a new envrionment following instruction in Pointcept

decode demeter parameter to 3d mesh of soybean

python decode.py --data_folder sample_params --sample_name 24_o --species soybean

python decode.py --data_folder sample_params --sample_name 08 --species ribes 

python decode.py --data_folder sample_params --sample_name 10008da --species maize

python decode.py --data_folder sample_params --sample_name 1 --species tobacco

python decode.py --data_folder sample_params --sample_name 02 --species rose

Raw 3D point clouds -> Demeter parameters
script_reconstruction/readme.md

Raw 3D point clouds -> L-system parameteres
third_party/CropCraft/readme.md

• sample data of soybean (2025-10-7)
• decoding (2025-10-7)
• editing tutorial (TBD)
• sample data of other species (2025-11-1)
• reconstruction from 3d point cloud (2025-10-8)
• building demeter representation from your own annotated 3d point cloud (TBD)
• learning leaf shape PCA from 2D leaf scanns (TBD)
• L-system baseline (2025-10-13)
• full soybean 3d dataset (2025-12-17)
• full soybean 2d image dataset (TBD)

This project is supported by NSF Awards #1847334 #2331878, #2340254, #2312102, #2414227, and #2404385. We greatly appreciate the NCSA for providing computing resources.

This code is released under the Academic Research License (Non-Commercial).
For commercial inquiries, please contact shenlong@illinois.edu.