Surf-D: Generating High-Quality Surfaces of Arbitrary Topologies Using Diffusion Models (ECCV 2024)

Surf-D: Generating High-Quality Surfaces of Arbitrary Topologies Using Diffusion Models
Project Page | Paper

This is an official implementation of Surf-D using PyTorch

We present Surf-D, a novel method for generating high-quality 3D shapes as Surfaces with arbitrary topologies using Diffusion models. Previous methods explored shape generation with different representations and they suffer from limited topologies and poor geometry details. To generate high-quality surfaces of arbitrary topologies, we use the Unsigned Distance Field (UDF) as our surface representation to accommodate arbitrary topologies. Furthermore, we propose a new pipeline that employs a point-based AutoEncoder to learn a compact and continuous latent space for accurately encoding UDF and support high-resolution mesh extraction. We further show that our new pipeline significantly outperforms the prior approaches to learning the distance fields, such as the grid-based AutoEncoder, which is not scalable and incapable of learning accurate UDF. In addition, we adopt a curriculum learning strategy to efficiently embed various surfaces. With the pretrained shape latent space, we employ a latent diffusion model to acquire the distribution of various shapes. Extensive experiments are presented on using Surf-D for unconditional generation, category conditional generation, image conditional generation, and text-to-shape tasks. The experiments demonstrate the superior performance of Surf-D in shape generation across multiple modalities as conditions.

Installation

We recommend to use Anaconda.

Create and activate a virtual environment.

conda env create -f environment.yaml
conda activate SurfD

cd meshudf
python3 setup.py build_ext --inplace

Download pretrained models

Download our pretrained models at google drive.

Generate from Diffusion:

Unconditional generation:

python -m sample.generate_uncond \
    --model_path pretrained_models/diffusion_uncond.pt \
    --output_dir ./outputs/uncond/ \
    --cond_mode no_cond \
    --ae_dir pretrained_models/ae_deepfashion3d.pt  \
    --num_samples 10 \
    --resolution 512

Sketch conditional generation:

python -m sample.generate_sketch \
    --model_path pretrained_models/diffusion_sketch.pt \
    --output_dir ./outputs/sketch_cond/ \
    --cond_mode sketch \
    --ae_dir pretrained_models/ae_deepfashion3d.pt \
    --sketch_path demo_images/sketch.png \
    --resolution 512

Image conditional generation:

python -m sample.generate_image \
    --model_path pretrained_models/diffusion_image.pt \
    --output_dir ./outputs/image_cond/ \
    --cond_mode img \
    --ae_dir pretrained_models/ae_pix3d.pt \
    --image_path demo_images/image.jpg \
    --mask_path demo_images/mask.jpg \
    --resolution 512

Text conditional generation:

python -m sample.generate_text \
    --model_path pretrained_models/diffusion_text.pt \
    --output_dir ./outputs/text_cond/ \
    --cond_mode text \
    --ae_dir pretrained_models/ae_text.pt  \
    --prompt "a dining chair" \
    --watertight --num_samples 10 \
    --resolution 512

Training

Prepare dataset

Down load DeepFashion3D dataset at DeepFashion3D, Pix3D dataset at Pix3D and ShapeNet dataset at ShapeNet.

Preprocess the dataset

cd AutoEncoder/encdc
python preprocess_udfs.py /path/to/data_root /path/to/output dataset_name

AutoEncoder Training:

cd AutoEncoder/encdc

python train_encdec.py ../cfg/xxx/xxx.yaml

Diffusion Training:

python train_diffcloth.py --cond_mode no_cond --save_dir xxx --overwrite --data_dir xxx --ae_dir xxx --log_interval 25 --save_interval 10000 --dataset deepfashion3d

Acknowledgement

Our code took reference from MDM, SDFusion, DrapeNet, MeshUDF, Stable Diffusion. We thank these authors for their great works and open-source contribution.

Citation

If you find this work useful for your research, please consider citing our paper:

@article{yu2023surf,
  title={Surf-D: High-Quality Surface Generation for Arbitrary Topologies using Diffusion Models},
  author={Yu, Zhengming and Dou, Zhiyang and Long, Xiaoxiao and Lin, Cheng and Li, Zekun and Liu, Yuan and M{\"u}ller, Norman and Komura, Taku and Habermann, Marc and Theobalt, Christian and others},
  journal={arXiv preprint arXiv:2311.17050},
  year={2023}
}

or

@inproceedings{yu2025surf,
  title={Surf-D: Generating High-Quality Surfaces of Arbitrary Topologies Using Diffusion Models},
  author={Yu, Zhengming and Dou, Zhiyang and Long, Xiaoxiao and Lin, Cheng and Li, Zekun and Liu, Yuan and M{\"u}ller, Norman and Komura, Taku and Habermann, Marc and Theobalt, Christian and others},
  booktitle={European Conference on Computer Vision},
  pages={419--438},
  year={2025},
  organization={Springer}
}