faster-particles 0.1.3

Point Proposal Network for particles images and related tools.

faster-particles

This package includes the following:

• Toydata generator
• LArCV data interface
• Pixel Proposal Network implementation using Tensorflow

Installation

You must install larcv2 and its own dependencies (ROOT, OpenCV, Numpy) in order to use LArCV data interface. To install larcv2:

git clone https://github.com/DeepLearnPhysics/larcv2.git
cd larcv2
source configure.sh
make

Then install faster-particles with Pip:

pip install faster-particles

Alternatively, you can also clone the source:

git clone https://github.com/Temigo/faster-particles.git
cd faster-particles

Usage

The following assumes you installed with pip. If you cloned the source, replace ppn with python bin/ppn.

To use toydata rather than LArCV data in the following sections, use option --toydata. LArCV data files can be specified with --data option. They can use regex, e.g. ppn_p[01]*.root.

Training

To train PPN on 1000 steps:

ppn train -o output/dir/ -l log/dir/ -d display/dir -n ppn -m 1000 --data path/to/data

To train the base network (currently only VGG available) on track/shower classification task:

ppn train -o output/dir/ -l log/dir/ -d display/dir -n base -m 1000

Inference

To run inference with a minimal score of 0.5 for predicted points:

ppn demo weights_file.ckpt -d display/dir/ -ms 0.5

More options are available through ppn train -h and ppn demo -h respectively.

Authors

K.Terao, J.W. Park, L.Domine