FOTBCD: A Large-Scale Building Change Detection Benchmark from French Orthophotos and Topographic Data

TL;DR

FOTBCD introduces a national-scale building change detection benchmark derived from France’s IGN BD ORTHO and BD TOPO data, spanning 28 departments with a 0.2 m resolution and split into 25 training and 3 held-out regions to probe geographic generalization. The dataset provides pixel-wise binary change masks (FOTBCD-Binary) and a smaller instance-level subset (FOTBCD-Instances) in COCO format, with careful quality control and spatial metadata in Lambert-93 coordinates. A fixed baseline, HybridSiam-CD, combines a frozen Vision Transformer semantic encoder with a CNN spatial branch to produce dense change maps, and is evaluated under a cross-domain setting against LEVIR-CD+ and WHU-CD. Results show that geographic diversity at the dataset level improves cross-domain robustness, with models trained on FOTBCD-Binary transferring more effectively to other datasets than models trained on geographically restricted benchmarks. The work provides publicly available benchmarks and annotations to foster ongoing research into geographically robust change detection, while acknowledging limitations such as national scope and building-centric labels.

Abstract

We introduce FOTBCD, a large-scale building change detection dataset derived from authoritative French orthophotos and topographic building data provided by IGN France. Unlike existing benchmarks that are geographically constrained to single cities or limited regions, FOTBCD spans 28 departments across mainland France, with 25 used for training and three geographically disjoint departments held out for evaluation. The dataset covers diverse urban, suburban, and rural environments at 0.2m/pixel resolution. We publicly release FOTBCD-Binary, a dataset comprising approximately 28,000 before/after image pairs with pixel-wise binary building change masks, each associated with patch-level spatial metadata. The dataset is designed for large-scale benchmarking and evaluation under geographic domain shift, with validation and test samples drawn from held-out departments and manually verified to ensure label quality. In addition, we publicly release FOTBCD-Instances, a publicly available instance-level annotated subset comprising several thousand image pairs, which illustrates the complete annotation schema used in the full instance-level version of FOTBCD. Using a fixed reference baseline, we benchmark FOTBCD-Binary against LEVIR-CD+ and WHU-CD, providing strong empirical evidence that geographic diversity at the dataset level is associated with improved cross-domain generalization in building change detection.

Figures (2)

• Figure 1: Geographic coverage of FOTBCD-Binary. Training departments (25) are shown in green; held-out departments (3) used for validation and testing are shown in blue.
• Figure 2: Qualitative visualization grid from FOTBCD.