Published on:
11 December 2023
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Ziyu Wan,
Despoina Paschalidou,
Ian Huang,
Hongyu Liu,
Bokui Shen,
Xiaoyu Xiang,
Jing Liao,
Leonidas Guibas
Avoids artifacts of prior mode-seeking 3D generation methods
Matches distributions adversarially rather than through sampling
Enables diverse 3D generation and reconstruction
Photorealistic and free-view synthesis
Photorealistic 3D generation from images
This paper proposes a new method to generate high-quality, diverse, and photorealistic 3D objects from a single input image and text description. It works by training a generative adversarial network to match the distribution of multi-view renderings to that of a pre-trained diffusion model. This avoids common issues like over-smoothing and saturation. The method enables applications like reconstruction, interpolation, and free-view synthesis.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper