PlantShade Dataset

Data for PlantShade: Predicting Plant Shadows for Lighting-Aware Robotic Agricultural Operation (IROS 2026).

A synthetic plant shade dataset rendered from a Helios + UE5 pipeline: aligned RGB, plant mask, shadow map, and depth captured from a fixed top-down (nadir) camera as a supplementary light orbits the canopy along a circular trajectory.

Contents

Each .zip is one scene. A scene folder holds per-growth-day subfolders N/ (RGB rgb_*.jpg, skeleton seg_*.png, depth depth_*.png) and N_shadow/ (the shadow-intensity render), with 100 supplementary-light positions per day.

• Species (4): tomato, soybean, sugar beet, strawberry
• Layouts (3): 1×1, 1×3, 3×5 (50 cm spacing)
• Growth stages: days 7 to 119
• Resolution: 1920 × 1080
• Light: 1.8 m height, 1.25 m radius, 100 circular positions

Scene name format: [<layout>]<species>_seed<n>_<start>_<interval>_<end>.zip.

Usage

from huggingface_hub import hf_hub_download
path = hf_hub_download(
    repo_id="xiao0o0o/PlantShade",
    filename="tomato_seed2_14_7_77.zip",
    repo_type="dataset",
)

Pipeline & model

Code (extract shadow, train, infer, photosynthesis) and the trained checkpoint:

• Website: https://darl-genai.github.io/PlantShade/
• Model: https://huggingface.co/xiao0o0o/PlantShade-ControlNet
• Data-collection simulator: https://github.com/ARLabXiang/AgriRoboSimUE5/releases/tag/plantshade

Citation

@inproceedings{da2026plantshade,
  title     = {PlantShade: Predicting Plant Shadows for Lighting-Aware Robotic Agricultural Operation},
  author    = {Da, Longchao and Liu, Xiaoou and Li, Xingjian and Xiang, Lirong and Wei, Hua},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year      = {2026}
}