Resources

Publicly available datasets

NOAA Puget Sound Nearshore Fish 2017-2018 - 77,739 images sampled from video collected on and around shellfish aquaculture farms in an estuary in the Northeast Pacific; 67,990 objects (fish and crustaceans) annotated on 30,384 images.

List of Marine/freshwater images

BRUVNet - An open-sourced dataset from baited remote underwater video of freshwater and marine fish images used for fisheries monitoring and research

Deep learning based Nephrops counter for demersal trawl fisheries

Publicly available models

MegaFishDetector v0

General purpose detector of fish shapes in images

KakaduFishAI

Multi-class detector for Australian fish

Training resources

Machine learning based image collection, annotation and classification

In this online course you will learn about:

  1. concepts of data science, machine learning (ML), computer vision, deep learning and Convolutional Neural Networks (CNNs)
  2. how to pre-process and pre-annotate images to accelerate your ML projects
  3. how to apply data augmentation techniques
  4. how to build an image classification model with your own or example data

The course uses freely available libraries and computing resources on Google Colaboratory

Scientific publications

A Scalable Open-Source Framework for Machine Learning-Based Image Collection, Annotation and Classification: A Case Study for Automatic Fish Species Identification. 2022 Catarina NS Silva, Justas Dainys, Sean Simmons, Vincentas Vienožinskis, Asta Audzijonyte. Sustainability 14 (21), 14324.

Preprint: A machine learning based image classification method to estimate fish sizes from images without a specified reference object. 2022. Catarina Nunes Soares Silva, Justas Dainys, Sean Simmons, Asta Audzijonyte. bioRxiv

Estimating catch rates in real time: Development of a deep learning based Nephrops (Nephrops norvegicus) counter for demersal trawl fisheries. 2023. Ercan Avsar, Jordan P. Feekings, Ludvig Ahm Krag. Frontiers in Marine Science

Conserve the open media ecosystem! Legal and ethical considerations when using online repositories for AI training in ecological research. 2023. Julian Lilkendey. Ecology Letters

This publication contributes to the discussion on the responsible use of media from online repositories for AI training. It offers best practices along the FAIR principles to help researchers use these resources ethically and legally. Particularly valuable is the comprehensive overview of Creative Commons (CC) licences and their implications for the use of CC licensed media in AI training, enabling researchers to understand and comply with the nuanced requirements of different licence types.

Herbivorous fish feeding dynamics and energy expenditure on a coral reef: Insights from stereo-video and AI-driven 3D tracking. 2024. Julian Lilkendey, Cyril Barrelet, Jingjing Zhang, Michael Meares, Houssam Larbi, Gérard Subsol, Marc Chaumont, Armagan Sabetian. Ecology and Evolution

The publication details an advanced methodological framework for fish detection, identification, and 3D tracking. Utilizing YOLOv5 for object detection, iNaturalist data for species identification, and DeepSORT for multi-object tracking, it describes and presents code for precise classification and tracking from stereo-video. Additionally, it includes code for data optimization and calculating rates of energy expenditure from fish movement. This integration of AI with conventional field methods enhances our understanding of energy dynamics in aquatic ecosystems.