We use the ShapeNetPart dataset (xyz, normals and labels) from here, which is prepared by PointNet++. Download the dataset and place it to data/shapenetcore_partanno_segmentation_benchmark_v0_normal/
.
You can specific your own data directory by changing the dataset
augment in train.py
.
The settings are similar as in our classification experiment. To train a model on this dataset (require 2 gpus for 2048 points input):
python train.py --gpu_idx 0 1
The models are saved every 10 epochs.
To evaluate a model from models/train/checkpoints/
:
python test.py --log train --checkpoint epoch_099.pkl
Also, you can save the predicted obj files (ground truth, prediction, difference):
python test.py --log train --checkpoint epoch_099.pkl --output ./results