Implementation of "Empowering Vector Graphics with Consistently Arbitrary Viewing and View-dependent Visibility". CVPR 2025, Highlight
git clone https://github.com/chenxinl/Dream3DVG.git
cd Dream3DVG
To set up the environment, please run:
chmod +x script/install.sh
sh script/install.sh
conda activate 3dvg
Remind to check the cuda version first (use nvcc -V), and modify the pytorch installation command in install.sh.
If you meet the gcc error when building or running pydiffvg, try to update gcc using:
conda install -c conda-forge gcc
Remind to modify the GPU arch in L48 of install.sh, take NVIDIA A100-sm_80 for example:
sed -i 's/set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -std=c++11")/set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -std=c++11 -gencode=arch=compute_80,code=sm_80")/' CMakeLists.txt
Directly run
sh train.sh
or
CUDA_VISIBLE_DEVICES='0' python svg_render.py x=dream3dvg seed=1 x.style='sketch' x.num_paths=32 "prompt='A benz car'" result_path='./workspace/3dvg/Benz' "x.camera_param.init_prompt='A benz car'"
CUDA_VISIBLE_DEVICES='0' python svg_render.py x=dream3dvg seed=1 x.style='iconography' x.num_paths=128 "prompt='A crab'" result_path='./workspace/3dvg/Crab' "x.camera_param.init_prompt='A crab'"
Thanks for the awesome repository:
DiffVG: https://github.com/BachiLi/diffvg
Pytorch-SVGRenderer: https://github.com/ximinng/PyTorch-SVGRender
LucidDreamer: https://github.com/EnVision-Research/LucidDreamer
3Doodle: https://github.com/changwoonchoi/3Doodle
Cite as below if you find this paper and repository are helpful to you:
@InProceedings{Li_2025_CVPR,
author = {Li, Yidi and Xiao, Jun and Lu, Zhengda and Wang, Yiqun and Jiang, Haiyong},
title = {Empowering Vector Graphics with Consistently Arbitrary Viewing and View-dependent Visibility},
booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
month = {June},
year = {2025},
pages = {18531-18540}
}