Advancing 3D Computer Graphics & Vision with Generative AI and Novel 3D Representation

This PhD project will investigate next-generation 3D computer vision and graphics by integrating state-of-the-art generative modelling techniques—in particular, diffusion models and flow-based generative models—with advanced 3D scene representations, such as Neural Radiance Fields (NeRF) and Gaussian Splatting (GS).

The overarching goal is to design accurate, efficient, and expressive generative frameworks for 3D scene understanding, reconstruction, and synthesis, with a strong emphasis on scalability and real-world applicability. By pushing the frontier of generative 3D modelling, the research is expected to make novel contributions to visual computing, deep generative learning, and foundation model development.

The project will address fundamental challenges in the following areas:

• 3D Scene Reconstruction and Novel View Synthesis: Leveraging diffusion and flow-based priors for robust reconstruction from sparse, noisy, or multi-modal input (e.g., RGB-D, multi-view video, text prompts).
• 3D Object Generation and Rendering: Developing controllable generative pipelines capable of producing geometry-aware and physically consistent 3D assets, with applications in gaming, digital twin creation, and content generation.
• Spatial-Temporal Modelling for 4D (Dynamic) Scenes: Extending static 3D generative models to capture temporal dynamics, enabling animation, video-driven 3D reconstruction, and generative simulation of human motion or complex interactions.
• Scalable Representations and Learning Algorithms: Exploring efficient training and inference strategies, including hierarchical representation and model distillation into compact foundation models, suitable for large-scale and real-time deployment.

Potential applications span medical imaging (3D/4D reconstruction and analysis), immersive media and VR/AR, autonomous robotics and navigation, and digital asset generation. The candidate will have the opportunity to contribute to both theoretical advancements (novel architectures, representation learning, and probabilistic modelling) and practical systems that broaden the impact of generative AI for 3D understanding and synthesis.

The candidates should have good knowledge of Math, Computer Vision and Deep Learning, and strong coding skill. Interested applicants are strongly advised to contact Dr. Jingjing Deng (jingjing.deng@bristol.ac.uk) to discuss the details.