BirdGAN

This repository includes data and code for the paper "Using Generative Adversarial Networks to Simulate Central-Place Foraging Trajectories"

Data

Trajectory datasets are contained in the data folder. These trajectories consist in foraging trips of distinct breeding seabirds species. During breeding period, seabirds need to feed and protect their chick. Therefore, they perform relatively short central-place foraging trips, where they leave their nest to get some food from the ocean and then they return to their nest to feed and protect their offspring.

Species	Country	Nb of trips
Sula variegata	Peru	78
Sula dactylatra	Brazil	50
Sula sula	Brazil	30

Code

Code for the paper are available in the code folder :

• 1_gan_selection_20_steps_SS.ipynb : consist in the comparison of different GAN architecture (CNN or LSTM) on a simplified dataset
• 2_gan_vs_hmm_200_steps_SD.ipynb : consist in the training of DCGAN with spectral regularization on the Sula dactylatra dataset
• 2_gan_vs_hmm_200_steps_SV.ipynb : consist in the training of DCGAN with spectral regularization on the Sula variegata dataset

Tutorial

A tutorial notebook is also proposed tutorial.ipynb. This notebook can be directly run on Google Colab by clicking here