This is the project repository of my team in the "Machine Learning for 3D Geometry" course at TUM.
A non-official PyTorch implementation for Scan2Mesh: From Unstructured Range Scans to 3D Meshes paper on ShapeNet dataset.
pip install -r requirements.txt
Install PyTorch3D from: https://github.com/facebookresearch/pytorch3d/blob/main/INSTALL.md
- Download ShapeNet dataset and store it in data folder
- Reduce the meshes in ShapeNet to at most 100 vertices using the notebook data/generation.py
- Train the vertix prediction model using the following command:
python train_vertix.py --config=configs/vertix_train.json - Train the edge prediction model using the following command:
python train_edge.py --config=configs/edge_train.json - Train the face prediction model using the following command:
python train_face.py --config=configs/face_train.json - Use the visualization notebook Visualize Results.ipynb to visualize the trained model on ShapeNet meshes.
