Back to Projects
SOILIE-3D
soilie3d.com
A 3D imagination engine for indoor scenes
A project for the Science of Imagination Lab (SOIL), Carleton University, Ottawa
A research-driven imagination engine that generates novel indoor 3D scenes from labeled RGB-D datasets. SOILIE-3D preprocesses object centroids, scene statistics, and triplet relationships, then uses those learned spatial patterns to place objects in Blender-rendered scenes. It visualizes a recency effect by fading earlier objects, and exports PNG frames, bird’s-eye views, GIF animations, and CSV scene data for analysis and presentation.
Highlights
- Preprocessing pipeline for RGB-D scenes, object centroids, triplet distances/angles, and per-scene statistics
- Blender-based rendering pipeline for imagined rooms, bird’s-eye views, and frame-by-frame animation output
- Research tooling for reproducible experiments plus a public site for showcasing generated results
Stack
PythonBlenderRGB-D datasetsSpatial triplets3D renderingNumPy + PandasPoint-cloud preprocessingPNG / GIF exportResearch tooling
Affiliations
Organizations and groups connected to this project.
Science of Imagination Lab (SOIL)
The Science of Imagination Lab (SOIL) at Carleton University (Ottawa) explores how imagination works and supports research software that models cognition and visual scene generation.