Overview

Code for the paper "Diffusion-based Semantic Image Synthesis from Sparse Layouts" (CGI 2023)

Check out our project page for more details.

Dependencies

See requirements.txt

Pre-trained Models

Original checkpoint used to evaluate the model in the paper can be downloaded from here. This ckpt takes 1024 steps to generate images, which can be very slow.

For a faster inference speed and slightly lower quality, please download a distilled checkpoint that only require 32 steps, from here.

How to Use

python sample.py --checkpoint "path/to/checkpoint"

Generated images will be saved in the "results" folder.

Citing

@inproceedings{Huang2023Sparse,
author={Huang, Yuantian and Iizuka, Satoshi and Fukui, Kazuhiro},
title={Diffusion-based Semantic Image Synthesis from Sparse Layouts},
booktitle={Computer Graphics International Conference},
year={2023},
organization={Springer},
}