Official implementation of HORIZON: High-Resolution Semantically Controlled Panorama Synthesis
A novel framework that generates high-quality, semantically-controlled 360-degree panoramas with minimal distortion
[Soon] Model weights release.
[01/2024] Code released.
[12/2023] Paper Accepted by AAAI 2024
[10/2022] Paper uploaded to arXiv.
Use Miniconda to setup environment for HORIZON. Setup the required environment by the following command:
conda env create -f env.yml
conda activate horizonPlease download our checkpoints from Azure Blob to run the following inference scripts. After download, make a ckpt folder follow the belown structure:
ckpt
├── CLIP
│ └── ViT-L-14.pt
├── dataset
│ └── pano #dataset folder
├── last.pth
└── vqg
└── VQGan16384F16.pth
bash script/inference.sh
When the inference finish, you'll get all the generated panorama under /ckpt/pano/horizon_mini/eval_visu/pred/
@misc{yan2022horizon,
title={HORIZON: A High-Resolution Panorama Synthesis Framework},
author={Kun Yan and Lei Ji and Chenfei Wu and Jian Liang and Ming Zhou and Nan Duan and Shuai Ma},
year={2022},
eprint={2210.04522},
archivePrefix={arXiv},
primaryClass={cs.CV}
}