Sequence Matters: Harnessing Video Models in 3D Super-Resolution

• DOI: 10.1609/aaai.v39i4.32458
• Publication Date: 2025-04-11T09:56:49Z
• Authors: Hyun-kyu Ko; Dongheok Park; Youngin Park; Byeonghyeon Lee; Juhee Han; Eunbyung Park
• Keywords: Sequence (biology)

3D super-resolution aims to reconstruct high-fidelity models from low-resolution (LR) multi-view images. Early studies primarily focused on single-image (SISR) upsample LR images into high-resolution However, these methods often lack view consistency because they operate independently each image. Although various post-processing techniques have been extensively explored mitigate inconsistencies, yet fully resolve the issues. In this paper, we perform a comprehensive study of by leveraging video (VSR) models. By utilizing VSR models, ensure higher degree spatial and can reference surrounding information, leading more accurate detailed reconstructions. Our findings reveal that remarkably well even sequences precise alignment. Given observation, propose simple practical approach align without involving fine-tuning or generating `smooth' trajectory trained over The experimental results show surprisingly algorithms achieve state-of-the-art tasks standard benchmark datasets, such as NeRF-synthetic Mip-NeRF 360 datasets.