Floor Plan CIS

Floor Plan CIS Dataset

Overview

This dataset contains 500 original, high-resolution floor plan images collected from real estate listings within the Russian Federation and CIS region. These images serve as the source domain data for the research project MitUNet and the associated article published in Machine Vision and Applications: https://doi.org/10.1007/s00138-026-01815-y.

Note: This dataset contains raw images without resizing or pre-applied augmentations. This allows researchers to apply their own preprocessing pipelines suitable for their specific model architectures.

Content & Challenges

The dataset features distinct regional architectural styles that challenge standard segmentation models:Texture-based Material Encoding: Differentiation between load-bearing walls (solid fills) and partition walls (hatching/textures).Non-Manhattan Geometry: Presence of curved walls and angled structures.Semantic Clutter: Heavy presence of dimension lines, text, and furniture outlines overlapping with structural elements.

Dataset Structure & Specifications

Image Format: JPG/PNG (Original resolution, variable dimensions).Preprocessing: None (Images are provided "as is").

Reproduction of MitUNet Results

To reproduce the results reported in the MitUNet paper, the following pipeline should be applied programmatically during training:Resize: Scale images to fixed 512x512 pixels.Augmentations: Apply geometric (rotation, perspective) and photometric (brightness, CLAHE) transformations. We recommend using the `Albumentations` library.Refinement: Ground truth masks in this dataset have already been cleaned (door/window openings subtracted) for better topological consistency.

Citation

If you utilize this dataset, please cite the official publication:

@Article{mitunet2026,
author = {Parashchuk, Dmitriy and Kaspshitskiy, Alexey and Karyakin, Yuriy},
journal = {Machine Vision and Applications},
title = {Enhancing floor plan recognition: a hybrid mix-transformer and U-Net approach for precise wall segmentation},
year = {2026},
number = {3},
pages = {53},
volume = {37},
doi = {10.1007/s00138-026-01815-y},
url = {https://doi.org/10.1007/s00138-026-01815-y}
}

A preprint version of this work is also available on arXiv: arXiv:2512.02413.