Project for EECS 542: Advanced Topics in Computer Vision.
Project Name: From 3D Gaussian Splatting to 3D Generative Model
Members: Jingxian Li, Zhiran Wang, Naichen Shi
We explore the challenge of generating 3D models from single or sparse-view 2D images, which is crucial for applications in augmented reality and heritage preservation. We reproduce the results from one-step and two-step methodologies (“Triplane Meets Gaussian Splatting" and "Large Multi-View Gaussian Model") that use advanced neural architectures and techniques like 3D Gaussian splatting for detailed model generation. We provide a series of evaluations that demonstrate significant advances in generative models that bridge the gap between 2D inputs and 3D outputs, enhancing both the sample efficiency and reconstruction fidelity.
