Completion

ShapeNet

Download the dataset for training and testing. The dataset is used by our paper on shape completion, which contains point clouds sampled from meshes of 8 categories in ShapeNet. The point clouds are in the format of ply, which can be visualized via viewers like meshlab. Clone the ocnn-pytorch repository, and enter the subdirectory projects, then run the following command.
```
python tools/ae_shapenet.py --run prepare_dataset
```
Run the following command to train the network. The training log and weights can be downloaded here.
```
python completion.py --config configs/completion.yaml
```
Run the following command to get the predictions on the testing dataest. The parameter following SOLVER.ckpt can be freely modified to test different trained weights. And the results are in the folder logs/completion/shapenet_eval.
```
python completion.py --config configs/completion.yaml               \
       SOLVER.run evaluate  SOLVER.alias eval                       \
       SOLVER.ckpt logs/completion/shapenet/checkpoints/00300.model.pth
```