[Official] Latent-L-systems:Transformer based Tree Generator🌲

Environment

conda env create -f environment.yml

Data

• .lstring files are uploaded to Google Drive. Please see the urls under data/lstrings.txt

Visualization

• Please refer to https://github.com/edisonlee0212/PlantArchitect
• I will add more explanations.

Steps

1. First, put a folder with *.lstring files on the same directory.

├── Acacia
│   ├── *.lstring
├── preprocessing.py

2. Run the Preprocessing file. It will do all the jobs. It will take about 2 hours using multi-processing with all CPUs.

python preprocessing.py

3. Training

• Set the correct file paths from Step2 in train.py Line#32, Line#36, Line#40, Line#41

python train.py

4. Inference

• Line#41, 43, 45 change the file path from the preprocessing. Line#46 update this to the model .pth file path.
• Line#47, the default is 5, which generate 5 lstring files. Ex) If you want to generate 100 files, set it to 100.
• Check def export_lstring_file function to update the save path for the lstring file.

python generate.py

If our paper has been helpful, we kindly ask that you cite it in your work. https://dl.acm.org/doi/10.1145/3627101

@article{10.1145/3627101,
author = {Lee, Jae Joong and Li, Bosheng and Benes, Bedrich},
title = {Latent L-Systems: Transformer-Based Tree Generator},
year = {2023},
issue_date = {February 2024},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {43},
number = {1},
issn = {0730-0301},
url = {https://doi.org/10.1145/3627101},
doi = {10.1145/3627101},
journal = {ACM Trans. Graph.},
month = {nov},
articleno = {7},
numpages = {16},
keywords = {geometric modeling, L-systems, neural networks}
}

Video