Home

Awesome

Introduction

This repository is a PyTorchDistributedDataParallel (DDP) re-implementation of the CVPR 2022 paper CrossPoint.

Preparation

Package Setup

  conda create -n crosspoint python=3.7.13
  codna activate crosspoint

  pip install torch==1.11.0+cu102 torchvision==0.12.0+cu102 --extra-index-url https://download.pytorch.org/whl/cu102
  pip install -r requirements.txt

pueue is a shell command management software, we use it for scheduling the model training & inference tasks, please refer to the official page for installation and basic usage. We recommend this tool because you can run the experiments at scale with its help thus save your time.

W&B Server Setup

We track the model training and fine-tuning with W&B tools. The official W&B tools may be slow and unstable since they are on remote servers, we install the local version by running the following command.

  sudo groupadd docker  # create `docker` group on Ubuntu
  sudo usermod -aG docker <username>  # add <username> to `docker` group to use docker, replace <username> with yours
  docker run --rm -d -v wandb:/vol -p 28282:8080 --name wandb-local wandb/local:0.9.41

If you do not have Docker installed on your computer before, referring to the official document to finish Docker installation on Ubuntu.

Pretrained Models

CrossPoint pretrained models with DGCNN feature extractor are available here.

Datasets

  1. Datasets are available here. Run the command below to download all the datasets (ShapeNetRender, ModelNet40, ScanObjectNN, ShapeNetPart) to reproduce the results.
    cd data
    source download_data.sh
    
  2. The directories of the downloaded datasets should be organized as follows.
    |- CrossPoint-DDP
    |---- data
    |-------- ShapeNetRendering
    |-------- modelnet40_ply_hdf5_2048
    |-------- ScanObjectNN
    |-------- shapenet_part_seg_hdf5_data
    
    The data directory is at the same level with models, scripts, etc.

Usage

Pre-train CrossPoint

./scripts/pt-cls.sh
./scripts/pt-partseg.sh

Downstream Tasks

1. Zero-shot 3D Object Classification

Run eval_ssl.ipynb notebook to perform linear SVM object classification in both ModelNet40 and ScanObjectNN datasets.

2. Few-Shot Object Classification

Refer scripts/fsl_script.sh to perform few-shot object classification.

3. Finetune CrossPoint for 3D Object Classification

./scripts/ft-mn.sh  # finetune on ModelNet40
./scripts/ft-so.sh  # finetune on ScanObjectNN

4. Finetune CrossPoint for 3D Object Part Segmentation

./scripts/ft-partseg.sh

Acknowledgements

The re-implementation is partially inspired by the following projects, thanks to their hard work.