Requirements

• python 3.10
• pytorch 2.1.0+cu121
• numpy 1.26.3
• pytorch-lightning 2.5.5
• polyscope 2.4.0
• tensorboard 2.20.0
• Run:

  pip install torch==2.1.0+cu121 \
    --index-url https://download.pytorch.org/whl/cu121 \
    numpy==1.26.3  

  pip install pytorch-lightning polyscope tensorboard

Download Data

• Download the ShapeNet dataset using this link:
  https://drive.google.com/drive/folders/1MMRp7mMvRj8-tORDaGTJvrAeCMYTWU2j?usp=sharing

Preprocess the Dataset

• Specify the parent folder where you stored the ShapeNet dataset in the preprocess section of configs/default.yaml.
• Run:

  python preprocess.py

Training

• Specify a category in the train section of configs/default.yaml.
  You can look at categ_to_id.json to see all available categories.
• Specify the location of the preprocessed dataset in configs/default.yaml.
• Run:

  python train.py

• You can use the following command to visualize training and validation losses.

  tensorboard --logdir logs

Sampling & Visualization

• In the sample section of configs/default.yaml, specify the path to the checkpoint of the trained model. By default, checkpoints of trained models are stored in the logs folder. You can also use the pre-trained checkpoint at the root level of the repository.
• Run:

  python sample.py

  It will display a polyscope interface, allowing you to visualize multiple sampled point clouds.