David Goldstein 1003477230
Tianxu An 1003099393
Runtian Wang 1003454102
Pretrained models are available for both the old TensorFlow DGCNN and the new simplified Pytorch DGCNN trained using permutation invariant cross entropy loss
The TensorFlow trained model is available at dgcnn/part_seg/train_results/trained_models
The Pytorch model is available at dgcnn/pytorch/checkpoints/exp/models
To evaluate the SAPIEN data using the TensorFlow model, go to dgcnn/part_seg and run
python3 test-Sapien.py
To evaluate the SAPIEN data using the Pytorch model, go to dgcnn/pytorch and run:
python3 test-sapien.py
Download the data for part segmentation by going to dgcnn/part_seg and running:
sh +x download_data.sh
To train, use at least 2 GPUs
Go to dgcnn/pytorch and run:
python3 main-perm-simple.py
Use the --epochs command to change the number of epochs, for example
python3 main-perm-simple.py --epochs 200