This repository contains the codes of our final project for EECS 598-012 in W21. Here is a link to our final report. It contains the implementation of PCN in PyTorch and an extension with self-attention layers and additinal classification loss. The self-attention layer is motivated by this paper.
Please download the Complete3D dataset from link and place it under the dataset folder. The dataloader is borrowed from the github of GRNet.
cd $./models/extensions/chamfer_dist
python setup.py install --user
The implementation of Chamfer distance is from the github of GRNet.
__C.DATASETS.COMPLETION3D.CATEGORY_FILE_PATH = '/home/mingyuy/pointcloud_completion/dataset/Completion3D.json'
__C.DATASETS.COMPLETION3D.PARTIAL_POINTS_PATH = '/home/mingyuy/pointcloud_completion/dataset/shapenet/%s/partial/%s/%s.h5'
__C.DATASETS.COMPLETION3D.COMPLETE_POINTS_PATH = '/home/mingyuy/pointcloud_completion/dataset/shapenet/%s/gt/%s/%s.h5'
| Average | Airplane | Cabinet | Car | Chair | Lamp | Sofa | Table | Watercraft | |
|---|---|---|---|---|---|---|---|---|---|
| PointFCAE | 17.52 | 5.71 | 20.18 | 8.23 | 20.22 | 30.60 | 15.15 | 26.54 | 13.53 |
| PCN | 16.64 | 5.13 | 21.02 | 8.15 | 19.68 | 26.33 | 14.28 | 26.51 | 12.05 |
| PCN_cla | 15.29 | 4.30 | 19.10 | 7.48 | 17.78 | 24.97 | 13.47 | 24.16 | 11.09 |
| PCN_trans | 13.65 | 3.68 | 18.77 | 7.10 | 16.65 | 19.50 | 12.34 | 22.03 | 9.12 |