The CSIRO Crown-of-Thorn Starfish Detection Dataset

Abstract

Crown-of-Thorn Starfish (COTS) outbreaks are a major cause of coral loss on the Great Barrier Reef (GBR) and substantial surveillance and control programs are underway in an attempt to manage COTS populations to ecologically sustainable levels. We release a large-scale, annotated underwater image dataset from a COTS outbreak area on the GBR, to encourage research on Machine Learning and AI-driven technologies to improve the detection, monitoring, and management of COTS populations at reef scale. The dataset is released and hosted in a Kaggle competition that challenges the international Machine Learning community with the task of COTS detection from these underwater images 1 .

A Comparative Analysis of Convolutional Neural Network and Vision Transformer Embeddings on a Novel Domain-Specific Task

“…It combines multi-head self-attention mechanisms with feedforward neural networks to capture global dependencies and contextual information within images, making it suitable for a wide range of computer vision tasks, including image classification, object detection, and semantic segmentation (Cuenat & Couturier, 2022). We used the pretrained MobileNetv2 and ViT-B16 to generate embeddings of the images in the COTS v NotCOTS Cropped Crown of Thorns Dataset (Liu et al, 2021). The "COTS v NotCOTS Cropped Crown of Thorns Dataset" serves as a valuable resource for researchers and practitioners interested in developing computer vision algorithms for detecting and classifying Crown of Thorns Starfish in underwater images.…”

“…The "COTS v NotCOTS Cropped Crown of Thorns Dataset" serves as a valuable resource for researchers and practitioners interested in developing computer vision algorithms for detecting and classifying Crown of Thorns Starfish in underwater images. Its focus on cropped images and binary classification makes it suitable for training and evaluating machine learning models for this specific task (Liu et al, 2021). The images of this dataset and its classes have not been introduced to our model at no point during their pretraining.…”

What's in the Black Box? The False Negative Mechanisms Inside Object Detectors

“…Given an image, object detectors identify target objects in terms of where they are and what they are. Detectors are useful tools for a range of robotics tasks -from object-centric mapping [1], to search-and-rescue missions [2], pest control in protected environments [3], and automated harvesting [4]. In each of these applications, the detector is one of the first components in a complex system, and its predictions can dictate the decisions made by follow-on components.…”