Easy-3Dfy: Object-Level Geometric Composition for Single-Image 3D Scene Construction
Ziang Zhang, Wenyou Yi, Zehan Wang, Jiachen Xu, Xintong Han, Chunchao Guo, Zhou Zhao
Zhejiang University · Tencent
Published October 23, 2026 · on the feed October 26, 2026 · ▲ 11 · arXiv 2610.32250
The abstract
Single-image 3D scene construction is a fundamental problem in 3D vision. Recent methods such as SAM-3Dfy enable end-to-end reconstruction of 3D assets and their scene layouts, but still struggle with fine-grained scene geometry and asset quality. Meanwhile, specialized 3D object generators provide detailed object geometry, while 3D foundation models offer strong and generalizable spatial understanding. We propose Easy-3Dfy, a compositional workflow that decouples object generation from scene layout estimation, allowing specialized models to fully exploit their respective strengths. An RPF-based aligner then bridges these capabilities by establishing geometric correspondence between object-local meshes and partial scene pointmaps, recovering object layout for scene composition. Without requiring complete scene-level supervision, the proposed workflow consistently outperforms SAM-3Dfy across multiple geometric metrics on Scene-Maker and 3D-FUTURE200. Our results demonstrate that explicit geometric composition can effectively combine high-quality object generation with strong spatial understanding, while enabling the workflow to benefit independently from advances in both 3D object generation and foundation models.
The Brief
2:06A short podcast summary of the paper. Written and voiced by AI.
Transcript
Easy-3Dfy. Object-Level Geometric Composition for Single-Image 3D Scene Construction. By Ziang Zhang and six others, from Zhejiang University and Tencent. Published on the twenty-third of October. No code repository is linked to it.
Building a 3D scene from a single image is a fundamental problem in 3D vision. Recent methods such as SAM-3Dfy reconstruct 3D assets and their scene layout end to end, but they still struggle with fine-grained scene geometry and with the quality of the assets. The end-to-end approach asks one system to get both the objects and their arrangement right. Meanwhile, two other kinds of model are each strong at one half of the job. Specialized 3D object generators provide detailed object geometry. 3D foundation models offer strong spatial understanding that generalizes.
Easy-3Dfy is a compositional workflow that separates those halves: object generation on one side, scene layout estimation on the other. Decoupling them lets each specialized model fully exploit its own strengths, instead of one end-to-end model having to be good at both.
The two halves still have to meet, and an aligner based on RPF is what bridges them. It establishes geometric correspondence between the object's local mesh and the partial pointmap of the scene. From that correspondence it recovers the object's layout, which is what the scene is then composed from. None of this requires complete scene-level supervision.
Across multiple geometric metrics, the workflow consistently outperforms SAM-3Dfy on two evaluations, Scene-Maker and 3D-FUTURE200. The comparison is with the end-to-end method the workflow was set against at the outset, and the measures are geometric ones. The authors take this as a demonstration that explicit geometric composition can effectively combine high-quality object generation with strong spatial understanding.
Scene-Maker and 3D-FUTURE200 are where the gains over SAM-3Dfy are measured, but the more lasting property is structural. Because objects and layout come from separate models, joined only through geometry, Easy-3Dfy can benefit independently from advances in 3D object generation and from advances in foundation models. Whichever side improves next, the workflow is built to take that improvement in.
More with an account
Listen to every day’s papers
Daily playlists, deep dives, quests for the papers, and quizzes on their key terms.
Join free