4D-LRM: Large Space-Time Reconstruction Model From and To Any View at Any Time

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Ma, Ziqiao, Chen, Xuweiyi, Yu, Shoubin, Bi, Sai, Zhang, Kai, Ziwen, Chen, Xu, Sihan, Yang, Jianing, Xu, Zexiang, Sunkavalli, Kalyan, Bansal, Mohit, Chai, Joyce, Tan, Hao
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912445120905216
author Ma, Ziqiao
Chen, Xuweiyi
Yu, Shoubin
Bi, Sai
Zhang, Kai
Ziwen, Chen
Xu, Sihan
Yang, Jianing
Xu, Zexiang
Sunkavalli, Kalyan
Bansal, Mohit
Chai, Joyce
Tan, Hao
author_facet Ma, Ziqiao
Chen, Xuweiyi
Yu, Shoubin
Bi, Sai
Zhang, Kai
Ziwen, Chen
Xu, Sihan
Yang, Jianing
Xu, Zexiang
Sunkavalli, Kalyan
Bansal, Mohit
Chai, Joyce
Tan, Hao
contents Can we scale 4D pretraining to learn general space-time representations that reconstruct an object from a few views at some times to any view at any time? We provide an affirmative answer with 4D-LRM, the first large-scale 4D reconstruction model that takes input from unconstrained views and timestamps and renders arbitrary novel view-time combinations. Unlike prior 4D approaches, e.g., optimization-based, geometry-based, or generative, that struggle with efficiency, generalization, or faithfulness, 4D-LRM learns a unified space-time representation and directly predicts per-pixel 4D Gaussian primitives from posed image tokens across time, enabling fast, high-quality rendering at, in principle, infinite frame rate. Our results demonstrate that scaling spatiotemporal pretraining enables accurate and efficient 4D reconstruction. We show that 4D-LRM generalizes to novel objects, interpolates across time, and handles diverse camera setups. It reconstructs 24-frame sequences in one forward pass with less than 1.5 seconds on a single A100 GPU.
format Preprint
id arxiv_https___arxiv_org_abs_2506_18890
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle 4D-LRM: Large Space-Time Reconstruction Model From and To Any View at Any Time
Ma, Ziqiao
Chen, Xuweiyi
Yu, Shoubin
Bi, Sai
Zhang, Kai
Ziwen, Chen
Xu, Sihan
Yang, Jianing
Xu, Zexiang
Sunkavalli, Kalyan
Bansal, Mohit
Chai, Joyce
Tan, Hao
Computer Vision and Pattern Recognition
Can we scale 4D pretraining to learn general space-time representations that reconstruct an object from a few views at some times to any view at any time? We provide an affirmative answer with 4D-LRM, the first large-scale 4D reconstruction model that takes input from unconstrained views and timestamps and renders arbitrary novel view-time combinations. Unlike prior 4D approaches, e.g., optimization-based, geometry-based, or generative, that struggle with efficiency, generalization, or faithfulness, 4D-LRM learns a unified space-time representation and directly predicts per-pixel 4D Gaussian primitives from posed image tokens across time, enabling fast, high-quality rendering at, in principle, infinite frame rate. Our results demonstrate that scaling spatiotemporal pretraining enables accurate and efficient 4D reconstruction. We show that 4D-LRM generalizes to novel objects, interpolates across time, and handles diverse camera setups. It reconstructs 24-frame sequences in one forward pass with less than 1.5 seconds on a single A100 GPU.
title 4D-LRM: Large Space-Time Reconstruction Model From and To Any View at Any Time
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.18890