About HerdNet

HerdNet is an open-source deep learning framework developed for the precise detection, identification and counting of African mammals in aerial imagery. Designed to assist researchers and conservationists, it leverages convolutional neural networks to analyze high-resolution aerial photographs, facilitating wildlife monitoring and ecological studies.

Purpose and Functionality: The primary aim of HerdNet is to automate the detection and enumeration of various mammal species in aerial images, thereby enhancing the efficiency and accuracy of wildlife population estimates. By inputting aerial photographs, users can obtain detailed information on animal locations, counts and species identification, which is crucial for ecological research and conservation efforts.

Key Features:

• Automated Detection and Counting: Utilizes advanced convolutional neural networks to detect, identify and count multiple mammal species in aerial imagery.
• Training and Testing Tools: Provides scripts (train.py and test.py) for initiating training sessions and evaluating model performance, allowing customization and improvement of detection capabilities.
• Inference Capabilities: Includes an infer.py tool to process new images and generate detection outputs, complete with visualizations and data exports.
• Visualization Support: Offers tools to visualize ground truth annotations and model detections, aiding in the assessment and refinement of model accuracy.
• Integration with PyTorch-Wildlife: Incorporated into the PyTorch-Wildlife platform, enabling users to load pre-trained HerdNet models and perform detections seamlessly.

Applications:

• Wildlife Conservation: Assists in monitoring animal populations, tracking changes over time, and making informed conservation decisions.
• Ecological Research: Facilitates studies on species distribution, habitat utilization, and behavioral patterns through efficient data processing.
• Aerial Survey Analysis: Enhances the analysis of aerial surveys by automating the detection and counting process, reducing manual effort and increasing accuracy.

Authors

Alexandre Delplanque

WEBSITE

Reference Publication