README.md

---
pretty_name: "WE3DS-small: A 576-Image Subset of WE3DS"
license: cc-by-4.0
language:
  - en
tags:
  - image
  - semantic-segmentation
  - agriculture
  - weed-detection
  - crop-farming
  - computer-vision
  - rgb-d
  - plant-species
task_categories:
  - image-segmentation
task_ids:
  - semantic-segmentation
size_categories:
  - n<1K
source_datasets:
  - synthgen/WE3DS
dataset_info:
  features:
    - name: image
      dtype: image
    - name: annotation
      dtype: image
    - name: date
      dtype: string
    - name: time
      dtype: string
    - name: latitude
      dtype: float64
    - name: longitude
      dtype: float64
    - name: weather
      dtype: string
    - name: wind
      dtype: string
    - name: seeding_date
      dtype: string
    - name: height_mm
      dtype: int32
  splits:
    - name: train
      num_examples: 344
    - name: test
      num_examples: 232
---

# WE3DS-small: A 576-Image Subset of WE3DS - https://zenodo.org/records/7457983

> **This is a curated subset of [synthgen/WE3DS](https://huggingface.co/datasets/synthgen/WE3DS). All credit goes to the original authors.**

## Overview

A lightweight **576-image subset** of the WE3DS dataset, designed for quick prototyping, demos, and experimentation without downloading the full 2,568-image dataset. Contains the same data fields and format as the full dataset.

## Original Authors

**Florian Kitzler**, **Norbert Barta**, **Reinhard W. Neugschwandtner**, **Andreas Gronauer**, **Viktoria Motsch**

University of Natural Resources and Life Sciences, Vienna (BOKU)

- **Full dataset**: [synthgen/WE3DS](https://huggingface.co/datasets/synthgen/WE3DS)
- **Original source**: [Zenodo (DOI: 10.5281/zenodo.7457983)](https://doi.org/10.5281/zenodo.7457983)
- **Paper**: [Sensors 2023, 23(5), 2713](https://doi.org/10.3390/s23052713)
- **License**: [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)

## Semantic Classes (19)

| ID | Class | Type |
|----|-------|------|
| 0 | void | background |
| 1 | soil | background |
| 2 | broad bean | crop |
| 3 | corn spurry | weed |
| 4 | red-root amaranth | weed |
| 5 | common buckwheat | crop |
| 6 | pea | crop |
| 7 | red fingergrass | weed |
| 8 | common wild oat | weed |
| 9 | cornflower | weed |
| 10 | corn cockle | weed |
| 11 | corn | crop |
| 12 | milk thistle | weed |
| 13 | rye brome | weed |
| 14 | soybean | crop |
| 15 | sunflower | crop |
| 16 | narrow-leaved plantain | weed |
| 17 | small-flower geranium | weed |
| 18 | sugar beet | crop |

## Dataset Structure

### Data Fields

- `image`: RGB image (PNG, 1600×1144)
- `annotation`: Segmentation mask (PNG, 1600×1144, pixel values = class IDs 0–18)
- `date`: Capture date (DD.MM.YYYY)
- `time`: Local capture time (HH:MM:SS)
- `latitude`: GPS latitude (WGS84)
- `longitude`: GPS longitude (WGS84)
- `weather`: Categorical — sunny, cloudy, or mixed
- `wind`: Categorical — light, medium, or strong
- `seeding_date`: Planting date (DD.MM.YYYY)
- `height_mm`: Estimated plant height in millimeters

### Data Splits

| Split | Images |
|-------|--------|
| Train | 344 (60%) |
| Test | 232 (40%) |

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("synthgen/WE3DS-small")

# Access a training example
example = dataset["train"][0]
image = example["image"]           # PIL Image (RGB)
mask = example["annotation"]       # PIL Image (segmentation mask)
```

For the full dataset (2,568 images), use:
```python
dataset = load_dataset("synthgen/WE3DS")
```

## Citation

If you use this dataset, **please cite the original authors**:

```bibtex
@article{Kitzler2023WE3DS,
  author    = {Kitzler, Florian and Barta, Norbert and Neugschwandtner, Reinhard W. and Gronauer, Andreas and Motsch, Viktoria},
  title     = {WE3DS: An RGB-D Image Dataset for Semantic Segmentation in Agriculture},
  journal   = {Sensors},
  year      = {2023},
  volume    = {23},
  number    = {5},
  pages     = {2713},
  doi       = {10.3390/s23052713}
}
```

```bibtex
@dataset{Kitzler2023WE3DSdata,
  author    = {Kitzler, Florian and Barta, Norbert and Neugschwandtner, Reinhard W. and Gronauer, Andreas and Motsch, Viktoria},
  title     = {WE3DS: An RGB-D image dataset for semantic segmentation in agriculture},
  year      = {2023},
  publisher = {Zenodo},
  version   = {v1},
  doi       = {10.5281/zenodo.7457983},
  url       = {https://doi.org/10.5281/zenodo.7457983}
}
```

## Acknowledgments

The original dataset was created as part of the **"DiLaAg – Digitalization and Innovation Laboratory in Agricultural Sciences"** project, funded by the Government of Lower Austria and the private foundation Forum Morgen.