The proposed HPGM consists of five components:
- text representation block
- graph conditioned layout prediction network (GC-LPN)
- floor plan post-processing
- language conditioned texture GAN (LCT-GAN)
- 3D scene generation and rendering
Figure: The overview framework of HPGM.
Python==3.7, PyTorch==1.8.1
We transform the dataset of Graph2Plan to the data format of Intelligent Home 3D. We split the data following Graph2Plan as well.
- Train GC-LPN
python main.py --cfg cfg/layout.yml --gpu '0'
- Test GC-LPN
python main.py --cfg cfg/layout_test.yml --gpu '0'
If you use any part of this code in your research, please cite the paper:
@inproceedings{chen2020intelligent,
title={Intelligent home 3d: Automatic 3d-house design from linguistic descriptions only},
author={Chen, Qi and Wu, Qi and Tang, Rui and Wang, Yuhan and Wang, Shuai and Tan, Mingkui},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={12625--12634},
year={2020}
}