InfiniteMesh: View Interpolation using Multi-view Diffusion for 3D Mesh Reconstruction
Keywords: View Interpolation, Multi-view Diffusion, 3D Mesh Reconstruction
Abstract: We present InfiniteMesh, a feed-forward framework for efficient high-quality image-to-3D generation with view interpolation. Recent advancements in Large Reconstruction Model (LRM) have demonstrated significant potential in extracting 3D content from multi-view images produced by 2D diffusion models. Nevertheless, challenges remain as 2D diffusion models often struggle to generate dense images with strong multi-view consistency, and LRMs often exacerbate this multi-view inconsistency during 3D reconstruction. To address these issues, we propose a novel framework based on LRM that employs 2D diffusion-based view interpolation to enhance the quality of the generated mesh. Leveraging multi-view images produced by a 2D diffusion model, our approach introduces an Infinite View Interpolation module to generate interpolated images from main views. Subsequently, we employ a tri-plane-based mesh reconstruction strategy to extract robust tokens from these multiple generated images and produce the final mesh. Extensive experiments indicate that our method generates high-quality 3D content in terms of both texture and geometry, surpassing previous state-of-the-art methods.
Supplementary Material: zip
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3567
Loading