Uncertainty SDF

Setup

git clone --recursive [email protected]:MMintLab/usdf.git
cd usdf
pip install -e .

Setup pointnet and pytorch-meta:

cd 3rd/pointnet.pytorch
pip install -e .
cd ../pytorch-meta
pip install -e .

Optional: Build the Manifold repo:

cd 3rd/Manifold
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j8

Data

Generate partial view dataset used for inference:

python scripts/dataset/render_dataset.py cfg/generative/test/test_v2.yaml test

Visualize:

python scripts/vis_dataset.py cfg/generative/test/test_v2.yaml test

Run Inference

Generate set of mesh predictions given partial views:

python scripts/generate.py cfg/generative/deepsdf_v1.yaml -d cfg/generative/test/test_v2.yaml -o out/generation/project/ours/ -v

Make Mesh Watertight

We use the Manifold repo to make sure meshes are watertight. Make sure you followed the instructions above to build the manifold repo.

1. Make single mesh watertight:

python scripts/make_mesh_watertight.py <path_to_input_mesh> <path_to_output_mesh>

2. Make a category of ShapeNet watertight:

python scripts/make_watertight_shapenet.py <path_to_input_category> <output_dir>