FireRedTeam 开源 InstanceAssemble：NeurIPS 2025 布局感知图像生成官方实现
GitHub - FireRedTeam/InstanceAssemble: Official implementation of "InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention" (NeurIPS 2025) · GitHub
FireRedTeam 开源了论文《InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention》的官方实现，该工作已被 NeurIPS 2025 收录。这是一个轻量级 Layout-to-Image 生成框架，可实现精确空间控制，并引入 DenseLayout 与 Layout Grounding Score（LGS）用于评估，在稀疏和密集布局上均达到 SOTA。权重可从 HuggingFace 下载，后续将发布文本控制版本、额外视觉控制版本并开源训练代码。
正文
当前显示原文，中文全文译文尚未就绪。
FireRedTeam; InstanceAssemble
Repository navigation
NotificationsYou must be signed in to change notification settings; Fork1 (1); Star46 (46)You must be signed in to star a repository
Repository files navigation
README
Official implementation of "InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention" (NeurIPS 2025).
Introduction
InstanceAssemble is a lightweight framework for Layout-to-Image generation that enables precise spatial control. We also introduce DenseLayout and Layout Grounding Score (LGS) for rigorous evaluation, where InstanceAssemble achieves state-of-the-art performance on both sparse and dense layouts.
TODO
Release textual control version; Release additional-visual control version; Open-source training code
2.1 Download from HuggingFace
You can either download the files manually, or run:
2.2 Directory Setup
All weights should be stored under ./pretrained. A correct setup looks like:
DenseLayout and Layout Grounding Score
1. Install GroundingDINO
For more details, please refer to GroundingDINO.
2. Generate Images by DenseLayout
3. Compute Layout Grounding Score
Citation
Contact