SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Go, Hyojun, Park, Byeongjun, Jang, Jiho, Kim, Jin-Young, Kwon, Soonwoo, Kim, Changick
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916692380090368
author Go, Hyojun
Park, Byeongjun
Jang, Jiho
Kim, Jin-Young
Kwon, Soonwoo
Kim, Changick
author_facet Go, Hyojun
Park, Byeongjun
Jang, Jiho
Kim, Jin-Young
Kwon, Soonwoo
Kim, Changick
contents Text-based generation and editing of 3D scenes hold significant potential for streamlining content creation through intuitive user interactions. While recent advances leverage 3D Gaussian Splatting (3DGS) for high-fidelity and real-time rendering, existing methods are often specialized and task-focused, lacking a unified framework for both generation and editing. In this paper, we introduce SplatFlow, a comprehensive framework that addresses this gap by enabling direct 3DGS generation and editing. SplatFlow comprises two main components: a multi-view rectified flow (RF) model and a Gaussian Splatting Decoder (GSDecoder). The multi-view RF model operates in latent space, generating multi-view images, depths, and camera poses simultaneously, conditioned on text prompts, thus addressing challenges like diverse scene scales and complex camera trajectories in real-world settings. Then, the GSDecoder efficiently translates these latent outputs into 3DGS representations through a feed-forward 3DGS method. Leveraging training-free inversion and inpainting techniques, SplatFlow enables seamless 3DGS editing and supports a broad range of 3D tasks-including object editing, novel view synthesis, and camera pose estimation-within a unified framework without requiring additional complex pipelines. We validate SplatFlow's capabilities on the MVImgNet and DL3DV-7K datasets, demonstrating its versatility and effectiveness in various 3D generation, editing, and inpainting-based tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2411_16443
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis
Go, Hyojun
Park, Byeongjun
Jang, Jiho
Kim, Jin-Young
Kwon, Soonwoo
Kim, Changick
Computer Vision and Pattern Recognition
Text-based generation and editing of 3D scenes hold significant potential for streamlining content creation through intuitive user interactions. While recent advances leverage 3D Gaussian Splatting (3DGS) for high-fidelity and real-time rendering, existing methods are often specialized and task-focused, lacking a unified framework for both generation and editing. In this paper, we introduce SplatFlow, a comprehensive framework that addresses this gap by enabling direct 3DGS generation and editing. SplatFlow comprises two main components: a multi-view rectified flow (RF) model and a Gaussian Splatting Decoder (GSDecoder). The multi-view RF model operates in latent space, generating multi-view images, depths, and camera poses simultaneously, conditioned on text prompts, thus addressing challenges like diverse scene scales and complex camera trajectories in real-world settings. Then, the GSDecoder efficiently translates these latent outputs into 3DGS representations through a feed-forward 3DGS method. Leveraging training-free inversion and inpainting techniques, SplatFlow enables seamless 3DGS editing and supports a broad range of 3D tasks-including object editing, novel view synthesis, and camera pose estimation-within a unified framework without requiring additional complex pipelines. We validate SplatFlow's capabilities on the MVImgNet and DL3DV-7K datasets, demonstrating its versatility and effectiveness in various 3D generation, editing, and inpainting-based tasks.
title SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2411.16443