AI Paper Atlas

Research map / 3D vision / 3D generation

Penetration-Free Compositional 3D Generation via Gaussian Surface Offset

Yilin Shao, Licheng Jiao, Lingling Li, Xu Liu, Fang Liu, Wenping Ma, Long Sun, Jiaxuan Zhao

ECCV 2026 · 3D vision › 3D generation · approach: Generative indoor scene synthesis · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Gaussian surface offset (GSO) (ablation)CompSAG · compositional text-to-3DB-VQA: 50.33 -> 57.28Improved
add Gaussian surface rectification (GSR) (ablation)CompSAG · compositional text-to-3DB-VQA: 52.19 -> 57.28Improved
use relation-aware SDS instead of unstructured SDS (ablation)CompSAG · compositional text-to-3DB-VQA: 54.47 -> 57.28Improved
add reference depth calibration (RDC) (ablation)CompSAG initialization · colliding volume ratioCVR (avg): 0.276 -> 0.239Improved
add tolerant collision loss (TCL) after RDC (ablation)CompSAG · colliding volume ratioCVR (avg): 0.239 -> 0.103Improved
replace TCL with Gaussian surface offset (ablation)CompSAG · colliding volume ratioCVR (avg): 0.103 -> 0.034Improved
fine-tune layout LLM (Q to Q*) (ablation)CompSAG · compositional text-to-3DB-VQA: 56.17 -> 57.28Improved
use human-adjusted layout (Q* to Q*+U) (ablation)CompSAG · compositional text-to-3DB-VQA: 57.28 -> 58.11Improved
fine-tuned layout (Q*) vs human-adjusted (Q*+U) (ablation)CompSAG · compositional text-to-3DCVR: 0.034 -> 0.038No clear effect
apply CompSAG on feedforward SAM3D initializationSAM3D · colliding volume ratioCVR: 0.276 -> 0.038Improved
apply CompSAG on feedforward MIDI initializationMIDI · colliding volume ratioCVR: 0.321 -> 0.038Improved
increase object count from 2 to 5CompSAG · runtimeminutes: 27 -> 34Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy