Text-to-3D content creation utilizing powerful text-to-image generation ability of Stable Diffusion and NeRF, supervised by Prof. Qiang Sun.
- Utilized knowledge distillation techniques to transfer knowledge from a teacher U-Net to a student U-Net for 2D diffusion models, achieving a 30% enhancement in 3D content generation speed
- Leveraged the advanced text-to-image generation capabilities of ControlNet and Stable Diffusion with conditional control for detailed and fine-grained 3D object creation, improving the stability of the generation process
- Incorporated 3D Gaussian Splatting in lieu of the traditional NeRF component, achieving a significant boost in generation speed and generated content quality
Yuxuan (Reacher) Zhang
Master of Science in Applied Computing (MScAC)
My research interests include diffusion models, 3D content generation, knowledge distillation and model compression.