A Three-Level Alignment Framework for Large-Scale 3D Retrieval and Controlled 4D Generation
Philip Xu, David Elizondo, Raouf Hamzaoui · 2025
Read A Three-Level Alignment Framework for Large-Scale 3D Retrieval and Controlled 4D Generation by Philip Xu, David Elizondo, Raouf Hamzaoui (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-22294.txt.
Description
Philip Xu introduces Uni4D, a single system for searching large 3D model collections with words and for generating controllable four-dimensional, time-varying assets. Both jobs depend on aligning three kinds of data: text descriptions, 3D models, and images. Uni4D builds that alignment in three levels on top of the Align3D 130 dataset.
A multi-head attention and search model first tightens text-to-3D retrieval. Multi-view alignment then links 3D models to their images, and image-to-text alignment supports generating temporally consistent 4D content. Tests show high-quality retrieval and controllable generation from the combined system. Xu presents Uni4D as a step toward practical dynamic multimodal understanding, where one alignment scheme serves both finding current assets and creating new animated ones.
Catalog record
| CREATOR | Philip Xu, David Elizondo, Raouf Hamzaoui |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-22294