MultiSet AIが異種モダリティ対応の「VPS Gen2」を公開 カメラ画像を3DマップタイルやGaussian Splat等の3Dデータと位置合わせ可能に。 既存のGISデータを直接活用でき、再撮影での再マップ作成が不要。 数年前の画像でも正確に特定可能。 詳細はリプ欄。

開発者ドキュメント

Aligning camera images with 3D map tiles and Gaussian splats without needing to remap is a real workflow shift. Traditional visual positioning systems required fresh scans of an environment, which meant every update to the underlying 3D data invalidated your reference imagery. Being able to reuse older imagery against new 3D representations breaks that cycle. The "images from several years ago can be accurately identified" claim is the most interesting part for OSINT. Historical photos with unknown locations become geolocatable if you can align them against current 3D city models. That's a workflow that used to require manual matching against Street View and hoping the buildings hadn't changed.