The source code for our paper "PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis" (CVPR 2024)
git clone
cd PLACE
conda env create -f environment.yaml
conda activate PLACE
Please follow the dataset preparation process in FreestyleNet.
The pre-trained models can be downloaded from GoogleDrive and should be put into the ckpt
folder.
After the dataset and pre-trained models are prepared, you may evaluate the model with the following scripts:
# evaluate on the ADE20K dataset
./run_inference_ADE20K.sh
# evaluate on the COCO-Stuff dataset
./run_inference_COCO.sh
For out-of-distribution synthesis, you just need to modify the ADE20K
or COCO
dictionary in the dataset.py
@article{lv2024place,
title={PLACE: Adaptive Layout-Semantic Fusion for Semantic Image Synthesis},
author={Lv, Zhengyao and Wei, Yuxiang and Zuo, Wangmeng and Kwan-Yee K. Wong},
journal={IEEE Conference on Computer Vision and Pattern Recognition},
year={2024}
}
Please send mail to cszy98@gmail.com