pumaguard 20.post32

No project description provided

PumaGuard

Introduction

Please visit http://pumaguard.rtfd.io/ for more information.

Get PumaGuard

GitHub Codespaces

If you do not want to install any new software on your computer you can use GitHub Codespaces, which provide a development environment in your browser.

Local Development Environment

A local development environment can be created by using the poetry tool, which can be installed with

sudo apt install python3-poetry

Run

poetry install

To install all of the necessary Python modules.

Running the scripts on colab.research.google.com

Google Colab offers runtimes with GPUs and TPUs, which make training a model much faster. In order to run the training script in Google Colab, do the following from the terminal:

git clone https://github.com/PEEC-Nature-Youth-Group/pumaguard.git
cd pumaguard
scripts/train.py --help

For example, if you want to train the model from row 1 in the notebook,

scripts/train.py --notebook 1

Running the server

The pumaguard-server watches a folder and classifies new files as they are added to that folder. Run with

poetry run pumaguard-server FOLDER

Where FOLDER is the folder to watch.

Training new models

For reproducibility, training new models should be done via the train script and all necessary data, i.e. images, and the resulting weights and history should be committed to the repository.

1. Get a TPU instance on Colab or run the script on your local machine.

2. Open a terminal and run

   git clone https://github.com/PEEC-Nature-Youth-Group/pumaguard.git
   cd pumaguard
   

3. Get help on how to use the script

   On Colab, run

   ./scripts/pumaguard --help
   ./scripts/pumaguard train --help
   

   On your local machine, run

   sudo apt install nvidia-cudnn
   poetry install
   poetry run pumaguard --help
   poetry run pumaguard train --help
   

4. Train the model from scratch

   ./scripts/pumaguard train --no-load --settings pumaguard-models/model_settings_6_pre-trained_512_512.yaml