PyTorch Implementation of PointNet paper.
The paper.
Blog post: here.
To train the model you have put the ModelNet10 or ModelNet40 datasets to dataset preprocessing that is explained in the blog post.
Once you've done it run the following to train:
python train.py
--did: Device id to train the model on (0)
--lr: Learning rate (0.001)
n_points: Number of sampled points (1024)
n_class: Number of classes (10)
batch_size: Mini batch size (32)
epochs: Number of epochs (100)
wd: Weight decay (0)
dropout: Dropout rate (0.3)