Data and Code to Reproduce Figures and Model Evaluation in "Using Transfer and Active Learning to Improve BirdNET Detection of Rare Nocturnal Birds"

This repository complements the paper titled “Using transfer and active learning to improve BirdNET detection of rare nocturnal birds from Passive Acoustic Monitoring in a Neotropical forest.” It contains the data used to train and evaluate models, the .tflite models themselves, and all the code and data necessary to reproduce the figures in the paper. For details on the workflow and methods, please refer to the companion repository: https://doi.org/10.5281/zenodo.18564530. 🗂️ Contents (as ZIPs) 1. BN_evaluation-and_retraining.zip Includes: 📁 train_set_0/:Initial training set used for transfer learning, with subfolders per class and the original training set tracker. 📁 model_test/:208 WAV files used for testing, plus 208 TXT annotation files used to evaluate model performance. 📁 models/:All models trained during the initial retraining (model 0) and through the active learning loop (models 1–9), saved as .tflite. Also includes: A .csv file with the number of training samples per model. A .csv with training parameters. A .txt with class labels used during model training. 2. to_reproduce_figures.zip Includes: 📁 data/:CSV files used for generating figures. 📁 code/:Jupyter Notebooks and one RMarkdown (.Rmd) file for figure generation. 📁 html/:HTML-rendered versions of the notebooks for easy viewing and reproducibility.