Skip to content

Equivariant point cloud (EPiC) classification network

Notifications You must be signed in to change notification settings

erikbuh/EPiC-classification

Repository files navigation

EPiC-classification

Permutation-equivariant binary classification network for variable-sized point clouds based on the discriminator used in the EPiC-GAN.

Reference: EPiC-GAN - Equivariant Point Cloud Generation for Particle Jets (arXiv:2301:08128)


Packages used:

  • pytorch
  • numpy
  • matplotlib
  • sklearn
  • comet_ml (for online logging)

Run the training for 10 epochs via:

#!/bin/bash
PARAMS=(
    --epochs 10
    --dataset_train TRAINING_SET
    --dataset_val VALIDATION_SET
    --dataset_test TEST_SET
    --logdir ./
)
python train.py "${PARAMS[@]}"

Find additional settings via python train.py --help or in the config.py file.


The dataloader expects the DATA_SETS to be in an .npz format with the data stored in the format [EVENTS, POINTS, FEATURES] (using zero-padding for variable-sized point clouds) and labels (0 or 1).

About

Equivariant point cloud (EPiC) classification network

Resources

Stars

Watchers

Forks

Languages