A transformer-based model that generates world land-use layouts given user controls.
Land-use refers to the way humans utilize and manage land, e.g., residential, farmland, and forest.
Read [Paper]
Step 1: Create a python environment using Anaconda or Miniconda:
conda create -n lutf python=3.10
conda activate lutf
Step 2: Install packages:
pip install -r requirements.txt
Run training:
python train.py --config configs/lutf.json
Run inference:
bash generate.sh
Download the evaluation data and put the .pkl files under data/evaluate/.
| Land-use | Value | Remark |
|---|---|---|
| Mask | 0 | To be generate by models |
| None | 1 | Not identified land |
| "Residential" | 2 | - |
| "Farmland" | 3 | - |
| "Industrial" | 4 | - |
| "Meadow" | 5 | - |
| "Grass" | 6 | - |
| "Retail" | 7 | - |
| "Recreation Ground" | 8 | - |
| "Forest" | 9 | - |
| "Commercial" | 10 | - |
| "Railway" | 11 | - |
| "Cemetery" | 12 | - |
@article{cheng2024learning,
title={Learning layout generation for virtual worlds},
author={Cheng, Weihao and Shan, Ying},
journal={Computational Visual Media},
year={2024},
publisher={Springer}
}