TomatoMAP is a novel dataset generated from our multi camera array based on findability, accessibility, interoperability, and reusability (FAIR). The data generation and annotation take two years with multiple domain experts. TomatoMAP includes three subsets, TomatoMAP-Cls, TomatoMAP-Det and TomatoMAP-Seg for 50 BBCH classification, 7 main area detection, and 10 classes instance segmentation for fine-grained phenotyping. The dataset has also unique 3D modeling potential for further research.
If you need any help, submit a ticket via GitHub Issues.
- TomatoMAP dataset is released under CC BY 4.0. Commercial use requires permission.
- TomatoMAP code space is released under Apache 2.0.
- 15.07.2025: Paper available on arXiv
- 18.07.2025 Full dataset release on e!DAL
- 18.07.2025 Submit to Nature
- 23.07.2025 Code repo public
- Claim e!DAL DOI
- TomatoMAP builder on Colab
- User friendly Web UI demo
- Update homepage
Please check code subfolder for more details. Or click:
Expand details
We suggest using conda for env management.
git clone https://github.com/0YJ/TomatoMAP.git
cd TomatoMAP
conda env create --file environment.yml
conda activate TomatoMAP
cd code
We use notebook as TomatoMAP builder (script version coming soon).
jupyter notebookUsing fine-tuned parameters for training your own model:
cp det/best_hyperparameters.yaml ./
# unzip TomatoMAP dataset you downloaded from our e!DAL repo under code/ folderThen follow the guide under TomatoMAP_builder.ipynb to finish the dataset setup.
TomatoMAP/
├── main.py # Main entry
├── README.md # Introduction
├── requirements.txt # Dependencies
│
├── avh/ # AI vs Human Analysis
│
├── seg/ # Segmentation package
│
├── cls/ # Classifier
│
├── det/ # Detection package
│ ├── TomatoMAP-Det.yaml # YOLo training settings
│ └── best_hyperparameters.yaml # Fine-tuned hyperparameters
│
├── trainers/ # Training modules
│ ├── cls_trainer.py # Classification trainer
│ ├── det_trainer.py # Detection trainer
│ └── seg_trainer.py # Segmentation trainer
│
├── datasets/ # Dataset handling
│ ├── cls_dataset.py # Classification dataset
│ └── seg_dataset.py # Segmentation dataset utilities
│
├── models/ # Model definitions
│ ├── cls_models.py # Classification models
│ └── seg_hooks.py # Segmentation training hooks
│
├── utils/ # Utility functions
│ ├── common.py # Common utilities
│ ├── visualization.py # Visualization tools
│ └── isat2coco.py # Format converter for Seg
│
└── outputs/ # Training outputs (created automatically)
├── cls/ # Classification results
├── det/ # Detection results
└── seg/ # Segmentation results
Train a classification model on TomatoMAP-Cls dataset:
# default training with MobileNetV3-Large
python main.py cls --data-dir ./TomatoMAP/TomatoMAP-Cls --epochs 100
# options
python main.py cls \
--data-dir ./TomatoMAP/TomatoMAP-Cls \
--model mobilenet_v3_large \
--epochs 100 \
--batch-size 32 \
--lr 1e-4 \
--img-size 640 640 \
--patience 5 \
--output-dir outputs/cls/experiment1Available models:
mobilenet_v3_large(default)mobilenet_v3_smallmobilenet_v2resnet18
Train a YOLO model on TomatoMAP-Det dataset:
# default training with YOLO11-Large
python main.py det --data-config ./det/TomatoMAP-Det.yaml --epochs 500
# options
python main.py det \
--data-config ./det/TomatoMAP-Det.yaml \
--model yolo11l.pt \
--epochs 500 \
--img-size 640 \
--batch-size 4 \
--patience 10 \
--device 0 \
--output-dir outputs/det/experiment1 \
--hyperparams ./det/best_hyperparameters.yamlTrain a Mask R-CNN FPN based model on TomatoMAP-Seg dataset:
# training
python main.py seg train \
--data-dir ./TomatoMAP/TomatoMAP-Seg \
--model COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yaml \
--epochs 100 \
--lr 0.0001 \
--batch-size 4 \
--patience 5
# evaluation
python main.py seg eval \
--data-dir ./TomatoMAP/TomatoMAP-Seg \
--model-path model_best.pth \
--output-dir outputs/seg
# visualization
python main.py seg vis \
--data-dir ./TomatoMAP/TomatoMAP-Seg \
--model-path model_best.pth \
--n 5 \
--output-dir outputs/seg
# dataset information
python main.py seg info --data-dir ./TomatoMAP/TomatoMAP-Seg
# analyze object size (small, big, middle)
python main.py seg analyze --data-dir ./TomatoMAP/TomatoMAP-SegAvailable models:
COCO-InstanceSegmentation/mask_rcnn_R_50_FPN_1x.yamlCOCO-InstanceSegmentation/mask_rcnn_R_50_FPN_3x.yamlCOCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yamlCOCO-InstanceSegmentation/mask_rcnn_X_101_32x8d_FPN_3x.yaml
TomatoMAP-Cls/
├── train/
│ ├── BBCH class1/
│ │ ├── img1.jpg
│ │ └── ...
│ └── BBCH class2/
│ └── ...
├── val/
│ └── ...
└── test/
└── ...
TomatoMAP-Det/
├── images
└── labels
TomatoMAP-Seg/
├── images/ # All images
│ ├── img1.JPG
│ └── ...
├── labels/ # All labels in COCO format
├── isat.yaml # Label and class configuration
└── img1.json
This project is powered by the de.NBI Cloud within the German Network for Bioinformatics Infrastructure (de.NBI) and ELIXIR-DE (Research Center Jülich and W-de.NBI-001, W-de.NBI-004, W-de.NBI-008, W-de.NBI-010, W-de.NBI-013, W-de.NBI-014, W-de.NBI-016, W-de.NBI-022), Ultralytics YOLO, Meta Detectron2, ISAT, and LabelStudio. Thanks to JetBrains for supporting us with licenses for their tools.
If you use TomatoMAP in your research and think our project is useful, please cite:
@misc{zhang2025tomatomultianglemultiposedataset,
title={Tomato Multi-Angle Multi-Pose Dataset for Fine-Grained Phenotyping},
author={Yujie Zhang and Sabine Struckmeyer and Andreas Kolb and Sven Reichardt},
year={2025},
eprint={2507.11279},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2507.11279},
}
@dataset{tomatomap,
title={TomatoMAP: Tomato Multi-Angle Multi-Pose Dataset for Fine-Grained Phenotyping},
author={Yujie Zhang and Sabine Struckmeyer and Andreas Kolb and Sven Reichardt},
journal={e!DAL-Plant Genomics and Phenomics Research Data Repository (PGP)},
year={2025}
}