edodema / Birdcalls

Birdcalls

A bird classification expert system based on the BirdCLEF2021 task and dataset, in the report only one model is described even though in the code we have more. This is due to two methods being developed but only the second one revealed to be computationally feasible.

N.b. I did not participate to the challenge nor did submit any model.

Structure

.
├── conf
│   ├── data
│   ├── demo
│   ├── hydra
│   ├── logging
│   ├── model
│   ├── optim
│   └── train
├── models
├── notebooks
├── report
│   └── images
└── src
    ├── common
    ├── demo
    ├── pl_data
    └── pl_modules

Requirements

• Python 3.9
• kaggle
• pip

Installation

First download the repository and place yourself in it.

$ git clone https://github.com/edodema/Birdcalls.git
$ cd Birdcalls

Then download dependencies, keep in mind that process could require one hour or more, depending on your connection. If one or more commans fail execute them separately, to get more information on the script run ./setup.sh -h.

$ chmod u+x setup.sh
$ ./setup.sh -dmo

Remember to configure the .venv environment, change absolute paths in .env.template and rename it to .env!

Demo

$ chmod u+x ./src/demo/run.sh
$ ./src/demo/run.sh