Type the Object, See the Asset

Type the Object, See the Asset is a research project exploring AI-powered 3D asset retrieval for future VR design workflows.

Rather than relying on exact file names or manual browsing, the system enables users to search large asset libraries using natural language descriptions. Each 3D asset is represented through a multi-view approach, where images rendered from multiple viewpoints are encoded into a shared embedding space. Text queries are embedded into the same space, allowing the system to retrieve semantically relevant 3D assets even when users provide only approximate descriptions.

To support immersive workflows, the retrieval model was deployed directly within a VR environment using ONNX, enabling real-time semantic search and visualization of results. By combining multi-view 3D representations with language-based retrieval, the project demonstrates a more intuitive approach to asset discovery in XR content creation environments.