Python 3.8
Pytorch 1.4.0
scikit-learn 0.24.1
torchvision 0.5.0
h5py 2.10.0
Run git clone command to download the source code
Download Modelnet40 dataset (The code will download itself, manually download and unzip to 'dataset' folder if error appears)
https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip
5-way-5-shot-5-task pointnet EGNN BCELoss
python train.py --device cuda --num_ways 5 --num_shots 5 --num_tasks 5 --emb_net pointnet --gnn_net egnn --loss bce
5-way-5-shot-5-task pointnet ours circle Loss
python train.py --device cuda --num_ways 5 --num_shots 5 --num_tasks 5 --emb_net pointnet --gnn_net ours --loss circle
5-way-5-shot-5-task pointnet ours circle Loss node_feats==64
python train.py --device cuda --num_ways 5 --num_shots 5 --num_tasks 5 --emb_net pointnet --gnn_net ours --loss circle --num_emb_feats 64
5-way-5-shot-5-task pointnet ours circle Loss gnn_layer==8 (82.2)
python train.py --device cuda --num_ways 5 --num_shots 5 --num_tasks 5 --emb_net pointnet --gnn_net ours --loss circle --num_graph_layers 8
5-way-5-shot-5-task dgcnn ours circle Loss
python train.py --device cuda --num_ways 5 --num_shots 5 --num_tasks 5 --emb_net ldgcnn --gnn_net ours --loss circle