Grapevine Segmentation Dataset

Grapevine-Seg is a semantic segmentation dataset designed for fine-grained visual understanding of grapevine scenes in agricultural and robotic perception applications. The dataset focuses on pixel-level annotation of grapevine-related elements captured in real-world vineyard environments, supporting research in precision agriculture, agricultural robotics, and computer vision–based crop monitoring.

The dataset is associated with the research presented in the paper:

“Grapevine-Seg: A Grapevine Segmentation Method Based on an Improved YOLACT.”
Published in OENO One.

DOI of the paper:
https://doi.org/10.20870/oenoone.2026.60.1.9628

The dataset provides the annotated images used to support the development and evaluation of the segmentation method described in the paper.

This dataset was originally released on OpenXLab at:
https://openxlab.org.cn/datasets/bandwa/Grapevine-Seg

To improve long-term accessibility, citation stability, and reproducibility, the dataset is now archived and published on Zenodo.

Data Collection

The images were collected under natural vineyard conditions, covering variations in:

• Viewing angles and distances

• Background complexity

These variations are intended to improve the robustness and generalization capability of segmentation models trained on this dataset.

Annotations

All images are annotated at the pixel level for semantic segmentation tasks. The annotations were manually labeled to ensure accuracy and consistency. The dataset is suitable for:

• Supervised semantic segmentation

• Training deep learning models for vineyard perception tasks

Dataset Structure

The dataset follows a clear and commonly used directory structure, facilitating straightforward integration with mainstream deep learning frameworks such as PyTorch and TensorFlow.

Typical use cases include:

• Semantic segmentation model training and evaluation

• Vision-based grapevine monitoring

• Agricultural robot perception and navigation

Licensing and Usage

The dataset is released for research and academic use. Users are encouraged to cite this Zenodo record when using the dataset in publications or derivative works.

Citation

If you use this dataset, please cite:

Dataset

Lingxin, B. Grapevine-Seg Dataset. Zenodo, Year. DOI: (Zenodo DOI: 10.5281/zenodo.18218165)

Associated Paper

Lingxin, B., et al. Grapevine-Seg: A Grapevine Segmentation Method Based on an Improved YOLACT. OENO One.
https://doi.org/10.20870/oenoone.2026.60.1.9628