Heterogeneous Point Set Transformers for Segmentation of Multiple View Particle Detectors

This repository contains code to train an HPST, and baselines like GAT and RCNN on NoVa Data for Multiple-view particle detector Segmentation, as seen on Heterogeneous Point Set Transformers for Segmentation of Multiple View Particle Detectors.

Setup

We recommend using conda for environment setup.

git clone https://github.com/erobl/hpst.git
cd hpst
conda create -n hpst python=3.10
conda activate hpst
pip install -r requirements.txt

Logging

We use WandB for logging. Please create a WandB project named "HPST" and use CLI login to use the code as is, or disable logging in the scripts.

Train HPST

python scripts/train.py --options_file "config/hpst/hpst_tune_nova.json" --name "{run_name}" --log_dir "runs" --gpus 4 

Train GAT

python scripts/train_gat.py --options_file "config/gnn/gat_tune_nova.json" --name "{run_name}" --log_dir "runs" --gpus 4

Train RCNN

python scripts/train_rcnn.py --options_file "config/rcnn/rcnn_tune_nova.json" --name "{run_name}" --log_dir "runs" --gpus 8