SCIRE is an interactive application for indexing, querying, and evaluating semantic sub-concepts in complex 3D scenes.
The application is built around a Godot/C# frontend, C# infrastructure services, Python-based descriptor extraction, and database-backed retrieval. It provides workflows for scene exploration, viewpoint indexing, semantic query formulation, retrieval result inspection, and benchmark evaluation.
SCIRE is a research prototype and is mainly intended for experimentation, evaluation, and development of scene-centric retrieval methods.
SCIRE allows users to:
- load and inspect complex 3D scenes,
- explore scenes through sampled viewpoints,
- index rendered viewpoints as retrievable sub-concepts,
- extract multimodal descriptors from rendered views,
- run text-based or benchmark-based retrieval queries,
- inspect ranked retrieval results in the scene,
- compare indexing and retrieval setups,
- run evaluation workflows for KIS and AMS tasks.
Unlike classical object retrieval tools, SCIRE works on sub-parts of larger 3D scenes. A result is therefore not just an object match, but a spatially grounded viewpoint or scene region that represents a semantic sub-concept.
The application consists of four main parts:
SCIRE
├── Godot / C# application
├── C# service and infrastructure layers
├── Python descriptors