Neural Visibility of Point Sets

This repository contains the implementation of Neural Visibility of Point Sets.

Neural Visibility of Point Sets
Jun-Hao Wang, Yiyang Tian, Baoquan Chen, Peng-Shuai Wang
Accepted by SIGGRAPH Asia 2025

• Neural Visibility of Point Sets
  • 1. Installation
  • 2. Training
  • 3. UI
  • 4. Citation

1. Installation

This code has been tested on Ubuntu 20.04 with 4 Nvidia 4090 GPUs (24GB memory).

1.1 Install Conda and create a Conda environment

conda create --name neuvis python=3.13
conda activate neuvis

1.2 Clone this repository and Install the required packages.

git clone https://github.com/Timekisser/NeuVis.git
conda install ocnn thsolver tqdm tensorboard

1.3 Install PyTorch=2.6.0 with conda according to the official documentation.

pip3 install torch torchvision torchaudio

2. Training

2.1 Download the ShapeNetV2 dataset and extract it to the data directory. You could also use other dataset, as long as the folder is organized as:

--ShapeNet
	--class1
		--model1
         --model_normalized.obj
         ...
		--model2
         --model_normalized.obj
         ...
		...
	--class2
	...

Then run the script:

python prepare_data.py --in_path [inpath] --out_path [outpath]

We provide the filelist of ShapeNetV2 in the data folder, which is defaultly used as the --out_path in the above command. If you want to use other datasets, please generate the filelist by yourself.

2.2 Run the training script

python train.py --config vis_shapenet.yaml SOLVER.gpu 0,1,2,3

2.3 Evaluate the model by:

python train.py --config vis_shapenet.yaml SOLVER.gpu 0, SOLVER.run test DATA.test.takes [numofpoints]

If you want to evaluate on each category on ShapeNet, you could run the script:

python script.py

See the script.py for more details. We also provide our pretrained model at Huggingface. You can download and put it in the logs/depth8/checkpoints folder, then run the script above to evaluate it.

3. UI

3.1 Download our pretrained model at Huggingface.

3.2 Install the required packages.

pip install -r requirements.txt

3.3 Prepare the data as .npz file.

We provide a script to convert the .obj files to .npz files. Run it by:

python sample.py --name [path-to-obj-file] --num_pts [number-of-points] 

3.4 Run the UI script.

python ui.py --model_path [path-to-model-weights]

You could drag and drop the .npz file to the UI window to visualize the result.

4. Citation

```bibtex
@inproceedings{wang2025neural,
  title={Neural Visibility of Point Sets},
  author={Wang, Jun-Hao and Tian, Yiyang and Chen, Baoquan and Wang, Peng-Shuai},
  booktitle={SIGGRAPH Asia},
  year={2025},
}