CS seminar title

Revolutionizing 3D scene editing: From neural fields to 3DGS with diffusion models and text-guided techniques

Speaker

Hasan Iqbal, Ph.D. candidate, Wayne State University

Abstract

This seminar presents a cohesive exploration of three innovative approaches to 3D scene editing, bridging the gap between textual prompts and high-quality 3D content. First, LatentEditor introduces a new framework for locally controlled, text-guided editing of neural fields. Using diffusion models, it embeds scenes into the latent space, allowing for faster and more accurate modifications than traditional methods. Next, Free-Editor offers a training-free solution for 3D scene editing by leveraging single-view editing to eliminate multi-view inconsistency, vastly improving speed and resource efficiency. Lastly, 3DEgo simplifies the 3D scene synthesis process from monocular videos using diffusion models, bypassing complex multi-stage workflows and providing faster, text-driven 3D scene generation. This seminar will highlight the technical innovations, practical applications, and performance gains these techniques offer, demonstrating how they push the boundaries of 3D scene editing across diverse datasets and scenarios.

Bio

I am a Ph.D. candidate at Wayne State University, specializing in cutting-edge research on Generative AI, Diffusion Models, 3D Computer Vision, Neural Radiance Fields (NeRFs), 3D Gaussian Splatting, and Virtual Reality. My work pushes the boundaries of 3D content generation and visualization, focusing on innovative techniques that blend AR/VR with advanced 3D modeling.

In addition to my research, I have co-authored several impactful publications, including work on anomaly detection using diffusion models, communication-efficient federated learning, and text-driven 3D scene editing. These contributions have been presented at top-tier conferences, including the European Conference on Computer Vision (ECCV).

Before my Ph.D., I worked as an Image Algorithm Engineer in Shanghai, tackling complex Computer Vision challenges like counterfeit detection and OCR with deep learning. I hold a Master’s degree from Tsinghua University, China and a Bachelor’s from the National University of Sciences and Technology, Pakistan both of which provided a strong foundation for my current work in 3D Computer vision and GenerativeAI.