Zero-shot text-to-3D-object generation with visual-geometry-grounded transformer

The generation of 3D objects from text has become a critical research area due to its potential to revolutionize fields like virtual reality, gaming, and industrial design. Current methods, however, face significant challenges, including the scarcity of high-quality, large-scale 3D datasets and the high computational cost of training models from scratch. This paper proposes a novel approach that leverages the power of pre-trained models to overcome these limitations. Our method integrates three key components: a text-to-image generation model to create a 2D image from a text prompt; a single-image-to-multi-view image model to synthesize consistent views from different angles; and a Visual Geometry Grounded Transformer to reconstruct a cohesive 3D representation from the multi-view images. This approach offers a more efficient and data-independent solution to the 3D generation problem, paving the way for more accessible and versatile content creation tools.