CanolaTrack

CanolaTrack is a curated dataset for leaf-level multi-object tracking (MOT) and detection from top-down RGB imagery of Brassica napus (canola) plants. Each sequence records a single plant over time; frames contain annotated bounding boxes with persistent leaf IDs for tracking.

• For baseline methods and a reference pipeline built on CanolaTrack, see LeafTrackNet (training, inference, and TrackEval integration) in our Github repo.

---

Dataset Summary

• Domain: Plant phenotyping (leaf-level analysis, time series)
• Modalities: RGB images (top-down)
• Use cases: Multi-object tracking (leaf IDs), detection, re-identification
• Content: Sequences of a single plant over days; each frame has MOT-style annotations
• Annotations: gt/gt.txt per sequence with frame, leaf_id, x, y, w, h (pixels)
• Extras: YOLOv10 proposals JSONs and LeafTrackNet model weightsfor reproducible tracking baselines

---

Repository Structure

CanolaTrack/ 
│  ├── train/
│  │   └── <plant_id>/
│  │         ├── gt/gt.txt # CSV: frame,id,x,y,w,h,,,*
│  │         └── img/{frame:08d}.jpg
│  └──val/
│     └── <plant_id>/
│            ├── gt/gt.txt
│            └── img/{frame:08d}.jpg
proposals/ # detection proposals for standardized benchmarking
│     ├── det_db_train.json
│     └── det_db_val.json
weights/ # detctors and tracker weights
      └── <files>

Supported Tasks and Benchmarks

• Multi-Object Tracking (MOT) at the leaf level
• Object Detection (per-frame leaf boxes)
• Leaf Segmentation (per-frame leaf masks)

---

How to Cite

Please cite the dataset and the accompanying papers:

@article{leaftracknet2027,
title = {LeafTrackNet: A deep learning framework for robust leaf tracking in top-down plant phenotyping},
journal = {Computers and Electronics in Agriculture},
volume = {256},
pages = {112417},
year = {2027},
issn = {0168-1699},
doi = {https://doi.org/10.1016/j.compag.2026.112417},
url = {https://www.sciencedirect.com/science/article/pii/S016816992601015X},
author = {Shanghua Liu and Majharulislam Babor and Christoph Verduyn and Breght Vandenberghe and Bruno Betoni Parodi and Cornelia Weltzien and Marina M.-C. Höhne},
}

CanolaTrack dataset© BASF SE 2025. This dataset may be freely used for non-commercial research and educational purposes.