batris-data/central-europe-rural-landscape-dataset-sample

SAMPLE VERSION (Preview Subset)

This repository contains the preview sample (45 images) of the Central European Rural Landscape Dataset (v1.0).

The full dataset (945 images, class-wise ZIP archives, full annotation set) is available separately under a commercial license.

Full version repository: 👉 https://huggingface.co/datasets/batris-data/central-europe-rural-landscape-dataset-full

For licensing inquiries: [email protected]

---

Central European Rural Landscape Dataset (v1.0)

Dataset Summary

Central European Rural Landscape Dataset (v1.0) is a curated image dataset designed for training and evaluation of machine learning models focused on landscape-level scene understanding in rural environments.

The dataset contains 945 images captured across diverse rural and natural settings in Central Europe, combining terrestrial photography and UAV (drone) imagery acquired in multiple seasons. The images represent a broad spectrum of landscape contexts, including mosaic rural landscapes, meadows and pastures, arable land, forest environments, water and wetland systems, mountainous terrain, and anthropogenic or technical landscape features.

Each image is assigned to one primary landscape class and may include supplementary tags describing additional visually significant elements such as vegetation types, terrain characteristics, water features, or human infrastructure. Annotation emphasizes dominant landscape context and spatial composition, rather than precise object-level detection.

The dataset is intended for image-based machine learning tasks such as scene classification, landscape context recognition, and analysis of spatial relationships between landscape elements. It does not contain georeferenced data or sensitive metadata and is suitable for research, development, and commercial applications under license.

Intended Use Cases

• Training and evaluation of landscape-level scene classification models in rural environments
• Recognition of dominant landscape context in complex outdoor scenes
• Learning visual relationships between landscape elements (e.g. forest–meadow, water–terrain, settlement–surrounding landscape)
• Development of AI models for environmental monitoring, land-use analysis, and rural landscape assessment
• Benchmarking computer vision models under real-world variability (seasonality, scale, viewpoint)

Dataset Scope and Design Principles

The dataset is designed to support higher-level interpretation of rural landscape scenes, emphasizing visual readability and dominant spatial context rather than administrative or cartographic boundaries.

Key design principles include:

• exactly one primary class per image,
• optional supplementary tags for additional significant elements,
• selective, visually driven annotation reflecting real-world ambiguity and gradual transitions between landscape types.

The dataset is not intended for object detection, segmentation, or precise inventory tasks, but for image-based landscape interpretation and classification.

A complete methodological documentation, including detailed class definitions, tagging rules, and annotation philosophy, is provided in the accompanying document: METHODOLOGY.md

Data Origin and Image Characteristics

• Number of images: 945
• Geographic context: Central Europe
• Image sources:
  • terrestrial photography
  • vertical and oblique UAV (drone) imagery
• Seasonal coverage: multiple seasons, including winter imagery
• Spatial scale: wide landscape views to local contextual scenes

Images were selected to reflect natural variability of rural landscapes rather than uniform acquisition conditions.

Technical Specifications of Image Data

• File format: JPEG (JPG)
• Color space: RGB
• Image content: natural-color landscape scenes without overlays or in-image annotations

Resolution and Orientation

Image resolution varies depending on source material and has not been artificially standardized:

• Minimum resolution: approximately 1596 × 898 pixels
• Maximum resolution: approximately 4032 × 2268 pixels

The original higher-resolution imagery was uniformly downscaled to improve storage efficiency, data transfer reliability, and practical usability in machine learning workflows, while preserving sufficient spatial detail for landscape-level analysis.

Images include both landscape and portrait orientations, with varying aspect ratios. Orientation and resolution reflect original acquisition conditions.

Metadata

All images were stripped of EXIF metadata prior to inclusion in the dataset. The dataset does not contain georeferenced information or links to coordinate reference systems.

Dataset Structure

Data Distribution Format

The dataset is distributed in two complementary forms:

• Full dataset: Provided as separate ZIP archives for each primary landscape class (C1–C7). Each archive contains all images belonging to the given class.
• Preview sample dataset: A small subset of images provided as unpacked JPEG files, intended for quick inspection of visual characteristics, class diversity, and annotation structure.

Annotations are provided separately as CSV files and apply to both the full dataset and the preview sample.

Preview / Sample Dataset

The preview sample consists of 45 images selected across all primary landscape classes to reflect basic visual diversity.
It is intended solely for exploratory purposes (e.g. browsing, testing data loading pipelines, reviewing annotation format).

The preview sample is not statistically representative and not suitable for model training or quantitative evaluation.

Primary Landscape Classes

The dataset is divided into seven primary classes (C1–C7) based on the dominant landscape character visible in each image:

• C1 – Mosaic Rural Landscape (224 images)
• C2 – Meadows and Pastures (395 images)
• C3 – Agricultural Landscape (Arable Land) (58 images)
• C4 – Forest Landscapes (188 images)
• C5 – Water and Wetland Environments (52 images)
• C6 – Mountain and Sub-mountain Terrain (12 images)
• C7 – Anthropogenic and Technical Landscape Features (16 images)

Each image belongs to exactly one primary class.

Detailed class definitions and classification principles are described in the methodological documentation.

Tagging Overview

In addition to the primary class, images may include supplementary tags that describe additional visually significant landscape or technical elements (e.g. forest edges, watercourses, infrastructure, terrain features).

Tags:

• are optional,
• do not duplicate the primary class,
• are applied selectively based on visual clarity and informational value.

A complete list of tags and their interpretation is provided in the methodological documentation.

Annotations

Annotations are provided as CSV files and are available separately for the full dataset and the preview sample.

• Full dataset annotations correspond to images stored inside the class-wise ZIP archives.
• Preview sample annotations correspond to the unpacked sample images provided for inspection.

Both annotation files follow the same schema and labeling conventions, enabling seamless transition from preview exploration to full dataset usage.

Annotations are provided in CSV format, linking each image file to:

• primary class identifier and name,
• optional supplementary tags (semicolon-separated).

Each image file is uniquely named and unambiguously linked to its annotation record.

Licensing and Usage

This dataset is provided under a commercial, non-exclusive license.

You may use the dataset for AI training, analytics, research, and commercial products or services.
You do not acquire ownership of the dataset, but all outputs and results created using it remain yours.

Redistribution, resale, sublicensing, or sharing of the dataset itself is not permitted.

Research use is permitted under the same license terms and does not allow redistribution of the dataset.

Full license terms are defined in the accompanying document: LICENSE.md

The same license terms apply equally to both the full dataset and the preview sample.

For commercial licensing inquiries, pricing, or custom terms, please contact: [email protected]

Versioning

• Dataset name: Central European Rural Landscape Dataset
• Version: v1.0
• Primary classes and tag definitions are frozen for this release.

Citation

If you use this dataset in academic work or publications, please cite it as:

Central European Rural Landscape Dataset (v1.0), BATRIS s.r.o., 2026.