Improving 3D Geometric Fidelity in Zero-Shot Text-to-3D Generation using Cross-View Correspondences

Leveraging cross-view correspondences computed from diffusion features to correct geometric flaws in NeRF-based text-to-3D generation models, improving the 3D fidelity of the output.