MMLandmarks: a Cross-View Instance-Level Benchmark for Geo-Spatial Understanding

Oskar Kristoffersen1,2, Alba R. Sánchez1, Morten R. Hannemose1,2, Anders B. Dahl1,2, Dim P. Papadopoulos1,2
1Technical University of Denmark, 2Pioneer Center for AI
Paper arXiv Code (coming soon) Data (coming soon)
Main paper figure

MMLandmarks: We present four distinct data modalities: ground-view images, aerial-view images, GPS coordinates, and textual descriptions, collected from 18,557 unique landmarks in the United States. Data sources are included alongside each modality.

Abstract

Geo-spatial analysis of our world benefits from a multimodal approach, as every single geographic location can be described in numerous ways (images from various viewpoints, textual descriptions, and geographic coordinates). Current geo-spatial benchmarks have limited coverage across modalities, considerably restricting progress in the field, as current approaches cannot integrate all relevant modalities within a unified framework.

We introduce the Multi-Modal Landmark dataset (MMLandmarks), a benchmark composed of four modalities: 197k high-resolution aerial images, 329k ground-view images, textual information, and geographic coordinates for 18557 distinct landmarks in the United States. The MMLandmarks dataset has a one-to-one correspondence across every modality, which enables training and benchmarking models for various geo-spatial tasks, including cross-view Ground-to-Satellite retrieval, ground and satellite geolocalization, Text-to-Image, and Text-to-GPS retrieval.

We demonstrate broad generalization and competitive performance against off-the-shelf foundational models and specialized state-of-the-art models across different tasks by employing a simple CLIP-inspired baseline, illustrating the necessity for multimodal datasets to achieve broad geo-spatial understanding.

Main paper figure

Collection Pipeline: Tags from OpenStreetMaps are used to collect Wiki-identifiers, ensuring that landmarks have a Wikipedia and Wikimedia Commons page. If both are available, we check that the longest edge of the landmark’s bounding box is smaller than 400 meters to keep an even size distribution across the dataset. Every resulting landmark has a Wikimedia Commons page (ground), a Wikipedia page (text), a box size and center (coordinates), and associated aerial imagery (satellite).

Dataset

The MMLandmarks dataset is a large-scale multimodal benchmark for geo-spatial understanding, containing four distinct data modalities: ground-view images, aerial-view images, GPS coordinates, and textual descriptions. The dataset encompasses 18,557 unique landmarks across the United States, with a total of 329k ground-view images, 197k aerial-view images, and corresponding textual descriptions and GPS coordinates.

Dataset comparison. Modality abbreviations: S - Satellite, G - Ground, T - Text, C - Coordinates, D - Drone. In the Scale column, the number in parentheses indicates the number of cities.
Task Dataset Year Train(G/S) Index(G/S) Instances Scale (Cities) Modalities Open-access License
Geo
localization		 IM2GPS 2008 6.4M/- - - Global G,C ✗ N/A
YFCC100M 2016 100M/- - - Global G,C ✓ Flickr TC
PlaNet 2016 126M/- - - Global G,C ✗ N/A
MP16 2017 4.7M/- - - Global G,C ✓ Flickr TC
OSV-5M 2017 5.1M/- - - Global G,C ✓ CC-BY-SA
Cross-View
Retrieval		 CVUSA 2015 35k/35k 8.8k/8.8k - USA(1) G,S ✓ Flickr TC
Vo. 2016 450k/450k 70k/70k - USA(11) G,S ✓ N/A
CVACT 2019 44k/44k 92k/92k - Australia(1) G,S ✓ N/A
Uni-1652 2020 11.6k/701 5.5k/1652 1652 72 Universities G,S,D ✓ N/A
VIGOR 2021 51k/44k 53k/46k - USA(4) G,S ✓ N/A
CV-Cities 2024 162k/162k 61k/61k - Global(16) G,S ✓ N/A
CVGlobal 2024 134k/134k - - Global(7) G,S ✓ N/A
Landmark
Retrieval		 R-Oxford 2018 - 5k + 1M/- 11 Oxford G ✓ Flickr TC/CC
R-Paris 2018 - 6k + 1M/- 11 Paris G ✓ Flickr TC/CC
GLDv1 2018 1.2M/- 1.1M/- 30k Global G ✗ Multiple
GLDv2 2020 4.1M/- 764k/- 200k Global G ✓ CC/Public-domain
MMLandmarks 2025 329k/197k 714k/100k 18,557 USA G,S,T,C ✓ CC/Public-domain

Examples

bbox examples

Visualization of the center GPS (green) and bounding boxes (purple) for the polygons associated with different landmarks.

BibTeX

@article{kristoffersen2025mmlandmarks,
  author    = {Kristoffersen, Oskar and Sanchez, Alba and Hannemose, Morten R. and Dahl, Anders B. and Papadopoulos, Dimitrios P.},
  title     = {MMLandmarks: a Cross-View Instance-Level Benchmark for Geo-Spatial Understanding},
  journal   = {arXiv preprint},
  year      = {2025},
  url       = {https://mmlandmarks.compute.dtu.dk}
}