Skip to main content
SiteIQ
Foundational AI Research

Research

We're advancing the science behind intelligent buildings. Through partnerships with leading universities, SiteIQ contributes to the open scientific community.

AI scene understanding visualization with holographic overlays on an indoor office environment
Featured Project

SceneTrans: Action-Aware Visual Scene Understanding

A large-scale paired-image dataset developed in collaboration with the University of Virginia to advance AI scene understanding and action reasoning. The dataset teaches AI to understand not just what changed between two images, but what action caused the change and how it could be executed.

14,000+
Annotated image pairs
6
Indoor scene environments
3
Action types (add · move · remove)
CC-BY-4.0
Open license
Method & Stack
NVIDIA Isaac Sim 5.1.0Physics-based renderingAutomated annotation pipelinePhotorealistic indoor scenesMulti-view paired capturesAction-aware annotations
From Paper to Platform

How foundational research becomes facility intelligence.

Every SceneTrans annotation is a building block of the Spatial Knowledge Graph that SiteIQ deploys in real buildings.

Scene perception

Action-aware models learn what changed in a space. Powers the live digital twin's ability to stay current as floors get reconfigured, assets are replaced, and tenants move.

Robot reasoning

Robots planning a route need to understand add, move, and remove events the same way a human supervisor would. SceneTrans teaches them that vocabulary at scale.

Capital planning

Continuous facility condition assessment depends on detecting change without an annual walk-through. Action-aware perception is the foundation.

Partnership

In collaboration with the University of Virginia

University of Virginia
Academic Partner

Led by Cong Shen, Associate Professor in the Department of Electrical and Computer Engineering at the University of Virginia. The collaboration brings together SiteIQ's expertise in large-scale data generation and simulation with UVA's deep research capabilities in multimodal AI and action-aware reasoning.

UVA Department of Electrical and Computer Engineering
Artifacts

Browse the work.

Code, data, and documentation — open to the research community.

Dataset (GitHub)

14,000+ paired indoor scene captures with action annotations.

Method overview

A short technical description of the annotation pipeline and the action-aware model architecture.

License

Released under CC-BY-4.0. Free for research and commercial use with attribution.

Citation

BibTeX entry for citing SceneTrans in your work.

BibTeX Citation
@misc{scenetrans2025,
  title  = {SceneTrans: Action-Aware Visual Scene Understanding},
  author = {Shen, Cong and SiteIQ Research},
  year   = {2025},
  url    = {https://github.com/ShenGroup/SceneTrans},
  note   = {Dataset, CC-BY-4.0}
}
Looking Ahead

What's next

We're continuing to invest in research that connects AI perception with real-world action in the built environment — extending paired-scene reasoning to richer modalities, longer action horizons, and sim-to-real deployment. More projects and publications coming soon.

Multi-modal scene groundingSim-to-real transferLong-horizon action sequences
Open to Collaboration

Working on something adjacent? Let's talk.

SiteIQ partners with academic researchers and applied AI groups on foundational problems in spatial intelligence. Reach out if there's an overlap.