SimpliMix

Official implementation of SimpliMix: A Simplified Manifold Mixup for Few-shot Point Cloud Classification (accepted by WACV 2024).

Installation

This project is built upon the following environment:

• Install Python 3.7
• Install CUDA 11.0
• Install PyTorch 1.10.2

The package requirements include:

• pytorch==1.10.2
• tqdm==4.63.1
• tensorboard==2.8.0

Datasets

• Download ModelNet40
• Download ModelNet40-C from Google Drive. In our experiments, we only use the LiDAR corruption pattern.
• Download ScanObjectNN
• Download ShapeNet

Train

Train protonet in the 5-way 1-shot setting on the ModelNet40 dataset:

python main.py --dataset modelnet40 --fs_head protonet --backbone dgcnn --k_way 5 --n_shot 1

Train protonet with simplimix in the 5-way 1-shot setting on the ModelNet40 dataset:

python main.py --dataset modelnet40 --fs_head protonet --backbone dgcnn_mm4 --k_way 5 --n_shot 1

Evaluate

python main.py --train False