IL3D: A Large-scale Indoor Layout Dataset for LLM-Driven 3D Scene Generation

🏠 Project Page

IL3D Dataset Usage Guide

This document provides an overview of the IL3D dataset.

Environment

We bulid this dataset with Ubuntu 22.04, Python 3.11, CUDA 11.8

Dataset Contents

The IL3D dataset consists of three main components:

3D-FRONT.zip: A 3D asset library containing 3D models for furniture and objects. HSSD.zip: A supplementary 3D asset library with additional high-quality 3D models. layout.zip: A scene layout library containing configuration files that define the arrangement of 3D assets in scenes.

Setup Instructions

Follow these steps to set up the environment and process the dataset: Unzip the Dataset FilesExtract the three compressed files using the following commands:

unzip 3D-FRONT.zip
unzip HSSD.zip
unzip layout.zip

This will create directories containing the 3D assets and scene layout configurations.

Set Up the Conda Environment

Create a new Conda environment named IL3D with Python 3.11:

conda create -n IL3D python=3.11
conda activate IL3D
pip install -r requirements.txt

Generate USDA Scene Models

Run the provided Python script to generate 3D scene models in USDA format:

python get_usda_rooms.py

This script processes the configuration files in the layout.zip archive and references the 3D assets from 3D-FRONT.zip and HSSD.zip. The USDA format requires absolute paths to the 3D assets, which is why this script generates the USDA files separately.