LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Jin, Haian, Jiang, Hanwen, Tan, Hao, Zhang, Kai, Bi, Sai, Zhang, Tianyuan, Luan, Fujun, Snavely, Noah, Xu, Zexiang
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917975443898368
author Jin, Haian
Jiang, Hanwen
Tan, Hao
Zhang, Kai
Bi, Sai
Zhang, Tianyuan
Luan, Fujun
Snavely, Noah
Xu, Zexiang
author_facet Jin, Haian
Jiang, Hanwen
Tan, Hao
Zhang, Kai
Bi, Sai
Zhang, Tianyuan
Luan, Fujun
Snavely, Noah
Xu, Zexiang
contents We propose the Large View Synthesis Model (LVSM), a novel transformer-based approach for scalable and generalizable novel view synthesis from sparse-view inputs. We introduce two architectures: (1) an encoder-decoder LVSM, which encodes input image tokens into a fixed number of 1D latent tokens, functioning as a fully learned scene representation, and decodes novel-view images from them; and (2) a decoder-only LVSM, which directly maps input images to novel-view outputs, completely eliminating intermediate scene representations. Both models bypass the 3D inductive biases used in previous methods -- from 3D representations (e.g., NeRF, 3DGS) to network designs (e.g., epipolar projections, plane sweeps) -- addressing novel view synthesis with a fully data-driven approach. While the encoder-decoder model offers faster inference due to its independent latent representation, the decoder-only LVSM achieves superior quality, scalability, and zero-shot generalization, outperforming previous state-of-the-art methods by 1.5 to 3.5 dB PSNR. Comprehensive evaluations across multiple datasets demonstrate that both LVSM variants achieve state-of-the-art novel view synthesis quality. Notably, our models surpass all previous methods even with reduced computational resources (1-2 GPUs). Please see our website for more details: https://haian-jin.github.io/projects/LVSM/ .
format Preprint
id arxiv_https___arxiv_org_abs_2410_17242
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias
Jin, Haian
Jiang, Hanwen
Tan, Hao
Zhang, Kai
Bi, Sai
Zhang, Tianyuan
Luan, Fujun
Snavely, Noah
Xu, Zexiang
Computer Vision and Pattern Recognition
Graphics
Machine Learning
We propose the Large View Synthesis Model (LVSM), a novel transformer-based approach for scalable and generalizable novel view synthesis from sparse-view inputs. We introduce two architectures: (1) an encoder-decoder LVSM, which encodes input image tokens into a fixed number of 1D latent tokens, functioning as a fully learned scene representation, and decodes novel-view images from them; and (2) a decoder-only LVSM, which directly maps input images to novel-view outputs, completely eliminating intermediate scene representations. Both models bypass the 3D inductive biases used in previous methods -- from 3D representations (e.g., NeRF, 3DGS) to network designs (e.g., epipolar projections, plane sweeps) -- addressing novel view synthesis with a fully data-driven approach. While the encoder-decoder model offers faster inference due to its independent latent representation, the decoder-only LVSM achieves superior quality, scalability, and zero-shot generalization, outperforming previous state-of-the-art methods by 1.5 to 3.5 dB PSNR. Comprehensive evaluations across multiple datasets demonstrate that both LVSM variants achieve state-of-the-art novel view synthesis quality. Notably, our models surpass all previous methods even with reduced computational resources (1-2 GPUs). Please see our website for more details: https://haian-jin.github.io/projects/LVSM/ .
title LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias
topic Computer Vision and Pattern Recognition
Graphics
Machine Learning
url https://arxiv.org/abs/2410.17242