Generative 3D Vision and Graphics
We develop generative models for creating, reconstructing, and controlling three-dimensional faces, human motion, objects, and scenes.
Explore themeLaboratory for Computational Vision and Intelligence
Advancing Generative Vision, Multimodal Perception, and Trustworthy AI
We develop generative, multimodal, and trustworthy AI systems that perceive, model, and understand complex visual environments.
- Established
- 2020
- Based at
- UBC Okanagan
- Research themes
- Three connected areas
Who we are
Research at the Intersection of Vision and Intelligence
The CVI Lab at the University of British Columbia develops advanced artificial-intelligence methods for visual computing, multimodal perception, and intelligent sensing. Our work connects foundational research in machine learning with practical challenges in three-dimensional generation, environmental understanding, signal analysis, and trustworthy AI.
Learn more about the labResearch themes
Our Research
We investigate complementary areas of artificial intelligence that connect generative modelling, real-world perception, and responsible deployment.
Multimodal Perception and Intelligent Sensing
We design learning-based systems that integrate images, video, audio, remote-sensing data, and other sensor signals to understand real-world environments.
Explore theme
Trustworthy and Explainable AI
We investigate artificial-intelligence systems that are interpretable, reliable, robust, and safe for use in complex real-world settings.
Explore themeSelected work
Featured Projects
Explore selected research projects from the CVI Lab.
TriSim
A remote-sensing retrieval framework that models image–image, image–text, and text–text similarities to identify and reduce false-negative training signals.
Universal Aesthetic Alignment
A position paper and benchmark study showing how image generators and reward models can override requests for unconventional, abstract, or deliberately anti-aesthetic imagery.
ArchitectHead
A 3D Gaussian head-avatar framework with continuous level-of-detail control, balancing visual quality against rendering cost without retraining.