pointnet

Implementation of PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

https://arxiv.org/pdf/1612.00593.pdf

Training Script

python3 train.py \
	--dataset-path=./ModelNet10 \
	--batch=32 \
	--lr=1e-3 \
	--epochs=100 \
	--checkpoint-every=1 \
	--evaluate-every=1 \
	--checkpoint-path=./checkpoint-big-100 \
	--jitter-augmentation=True \
    --rotation-augmentation=False

Inference Script

python inference.py \
	--checkpoint-path=./checkpoint-big-100/epoch-96.pt \
	--obj-path=./ModelNet10/bathtub/train/bathtub_0001.off

Training

We trained PointNet on ModelNet10 for 100 epochs with a batch size of 32 and a learning rate of 1e-3 using the Adam optimizer. We sample 1024 points from the surface of the mesh, normalize them to an unit sphere, and add gaussian noise with a mean of 0 and standard deviation of 0.02.

Classification Accuracy (Overall + Per Class)

Overall	Bathtub	Bed	Chair	Desk	Dresser	Monitor	Nightstand	Sofa	Table	Toilet
90.2%	92.0%	95.0%	98.0%	88.4%	91.9%	96.0%	55.8%	100.0%	85.0%	96.0%