Game4Loc: UAV Geo-Localization Benchmark from Synthetic Game Data

Game4Loc is a UAV geo-localization benchmark created by Yux1angJi from synthetic game data, featured as an AAAI 2025 Oral presentation. The dataset provides visual imagery and geographic coordinates for visual place recognition and localization tasks, updated as of October 2025.

Use Cases

• Training visual place recognition models using synthetic UAV-view images
• Evaluating cross-view geo-localization algorithms against game-derived ground truth
• Benchmarking image-to-image retrieval for UAV navigation systems

Strengths

• AAAI 2025 Oral publication status
• Apache-2.0 open-source license
• Synthetic game data allows for precise ground truth geographic labels

Limitations

• Synthetic data may have a domain gap when applied to real-world UAV sensors
• Geographic scope is limited to a virtual game environment

Provenance

Source

AAAI 2025 Oral Paper: Game4Loc

Collection Method

Synthetic data generation from a game engine

Freshness

Updated October 2025; reflects AAAI 2025 research state.

Geography

Virtual environment