RumexLeaves

RumexLeaves is a image dataset with fine-grained annotation of Rumex obtusifolius weeds. For each indivudal leaf, we provide a pixel-level segmentation mask as well as a keypoint-guided polyline along stem and major vein of the leaf. The dataset includes 809 images with 7747 annotations, while it is differntiated between two types of datapoints: (1) iNaturalist datapoints have been downloaded from the plant publisher iNaturalist and (2) RoboRumex that have been collected with a Husky Robot platform. Both variants originate from real-world settings. The following table gives an overview of the dataset.

	Total	iNaturalist	RoboRumex
# Images	809	690	119
# Leaves w stem	3460	3250	210
# Leaves wo stem	4287	3916	371

Example Images

iNaturalist Samples

RoboRumex Samples

Getting started: Pytorch Dataset Class

Download data

wget https://data.dtu.dk/ndownloader/files/41521812

Install dependencies

pip install -r requirements.txt

The Pytorch Datasets allows an easy entrypoint to work with the dataset. To visualize some example images, please run.

python rumex_leaves/visualize_img_data.py --data_folder <path-to-your-extracted-RumexLeaves-folder> --num_images <number-of-images-to-display> --datapoint_type <iNaturalist/RoboRumex>

Citation

If you find this work useful in your research, please cite:

@article{fine_grained_2023,
author = {Güldenring, Ronja and Anderse, Rasmus Eckholdt and Nalpantidis, Lazaros},
title = {Fine-grained Leaf Analysis for Efficient Weeding Robots},
journal = {tba},
year = {2023}
}