Repurposing 3D Generative Model for Autoregressive Layout Generation
Summary
LaviGen is a framework that repurposes 3D generative models for autoregressive 3D layout generation, using an adapted 3D diffusion model with dual-guidance self-rollout distillation to achieve 19% higher physical plausibility and 65% faster computation than state-of-the-art methods on the LayoutVLM benchmark.
View Cached Full Text
Cached at: 04/20/26, 08:27 AM
Paper page - Repurposing 3D Generative Model for Autoregressive Layout Generation
Source: https://huggingface.co/papers/2604.16299
Abstract
LaviGen introduces a 3D layout generation framework that uses an adapted 3D diffusion model with dual-guidance self-rollout distillation for improved efficiency and spatial accuracy.
We introduce LaviGen, a framework that repurposes3D generative models (https://huggingface.co/papers?q=3D%20generative%20models)for3D layout generation (https://huggingface.co/papers?q=3D%20layout%20generation). Unlike previous methods that infer object layouts from textual descriptions, LaviGen operates directly in the native 3D space, formulating layout generation as anautoregressive process (https://huggingface.co/papers?q=autoregressive%20process)that explicitly modelsgeometric relations (https://huggingface.co/papers?q=geometric%20relations)andphysical constraints (https://huggingface.co/papers?q=physical%20constraints)among objects, producing coherent and physically plausible 3D scenes. To further enhance this process, we propose an adapted3D diffusion model (https://huggingface.co/papers?q=3D%20diffusion%20model)that integrates scene, object, and instruction information and employs adual-guidance self-rollout distillation (https://huggingface.co/papers?q=dual-guidance%20self-rollout%20distillation)mechanism to improve efficiency and spatial accuracy. Extensive experiments on theLayoutVLM benchmark (https://huggingface.co/papers?q=LayoutVLM%20benchmark)show LaviGen achieves superior3D layout generation (https://huggingface.co/papers?q=3D%20layout%20generation)performance, with 19% higher physical plausibility than the state of the art and 65% faster computation. Our code is publicly available at https://github.com/fenghora/LaviGen.
View arXiv page (https://arxiv.org/abs/2604.16299)View PDF (https://arxiv.org/pdf/2604.16299)Project page (https://fenghora.github.io/LaviGen-Page/)GitHub10 (https://github.com/fenghora/LaviGen)Add to collection (https://huggingface.co/login?next=%2Fpapers%2F2604.16299)
Get this paper in your agent:
hf papers read 2604\.16299
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2604.16299 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2604.16299 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2604.16299 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollection (https://huggingface.co/new-collection)to link it from this page.
Similar Articles
Accelerating Disaggregated RL for Visual Generative LLMs with Diffusion-Based Parallelism and Trainer-Assisted Generation
This paper introduces DigenRL, a disaggregated RL framework for diffusion-based generative LLMs that uses generation-axis pipeline parallelism and trainer-assisted generation to improve throughput by 1.56-2.10x over existing systems.
Generative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards
This paper introduces a text-based approach for generative floor plan design that fine-tunes a large language model with reinforcement learning and verifiable rewards to improve adherence to topological and numerical constraints, achieving significant improvements over existing methods.
Native and Compact Structured Latents for 3D Generation
This paper introduces O-Voxel, a new sparse voxel representation for 3D generative modeling that efficiently handles complex topologies and appearance, and trains large-scale flow-matching models with 4B parameters to achieve state-of-the-art generation quality.
Scaling Novel Graph Generation via Lightweight Structure-Guided Autoregressive Models
Researchers propose a lightweight autoregressive framework for graph generation that uses structure-guided topological ordering to achieve near log-linear complexity, addressing scalability and novelty limitations of existing diffusion and autoregressive methods. The approach supports both LSTM and Mamba-style backbones and shows improved novelty while maintaining validity and uniqueness on molecular and non-molecular benchmarks.
Language Generation as Optimal Control: Closed-Loop Diffusion in Latent Control Space
This paper reformulates language generation as a stochastic optimal control problem, addressing limitations of autoregressive and diffusion models, and proposes a closed-loop diffusion method in latent control space using Flow Matching, achieving high-fidelity generation and efficient parallel sampling.