Datasets:

retgenai
/

FOTBCD-Binary

license: cc-by-nc-sa-4.0
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: image_id
      dtype: string
    - name: image_before
      dtype: image
    - name: image_after
      dtype: image
    - name: label
      dtype: image
    - name: dep
      dtype: string
    - name: epsg
      dtype: int32
    - name: patch_size
      list: int32
    - name: patch_bounds
      dtype: string
    - name: year_before
      dtype: int32
    - name: year_after
      dtype: int32
  splits:
    - name: train
      num_bytes: 19237579701
      num_examples: 25908
    - name: validation
      num_bytes: 778303470
      num_examples: 984
    - name: test
      num_bytes: 771375873
      num_examples: 979
  download_size: 20780292545
  dataset_size: 20787259044
task_categories:
  - mask-generation
  - image-segmentation
tags:
  - Geospatial
  - Remote Sensing
  - Building Change Detection
size_categories:
  - 10K<n<100K

FOTBCD-Binary

A large-scale building change detection benchmark from French orthophotos and topographic data.

Dataset Description

Property	Value
Departments	28 (25 train / 3 eval)
Image pairs	~28k
Patch size	512×512
Resolution	0.2m
Annotation	Binary mask

Splits

Split	Examples
train	~26k
val	~1k
test	~1k

Features

Field	Type	Description
`image_id`	string	Unique identifier
`image_before`	Image	Before image (RGB)
`image_after`	Image	After image (RGB)
`label`	Image	Binary mask (0=unchanged, 255=changed)
`dep`	string	French department code
`epsg`	int	Coordinate reference system
`patch_bounds`	string	Geographic bounds (JSON)
`year_before`	int	Year of before image
`year_after`	int	Year of after image

Usage

from datasets import load_dataset

ds = load_dataset("retgenai/FOTBCD-Binary")

example = ds["train"][0]
example["image_before"].show()
example["image_after"].show()
example["label"].show()

License

Dataset: CC BY-NC-SA 4.0
Source: BD ORTHO / BD TOPO by IGN under Licence Ouverte 2.0

Citation

If you use FOTBCD in your research, please cite our paper:

@misc{moubane2026fotbcd,
      title={FOTBCD: A Large-Scale Building Change Detection Benchmark from French Orthophotos and Topographic Data},
      author={Abdelrrahman Moubane},
      year={2026},
      eprint={2601.22596},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2601.22596},
}

Contact

Commercial licensing (FOTBCD-220k): info@retgen.ai