Squidle+ (squidle.org) is a marine image data management, discovery and annotation platform underpinning the IMOS Understanding Marine Imagery sub-facility. It provides access to millions of images and annotations.
Squidle+ features a user group framework, facilitating sharing and collaboration between users and also with external algorithms. Algorithms are set up as "users" of the system that interact through the comprehensive API backend. This architecture enables a variety of automated processing pipelines, connecting independent machine learning (ML) researchers to real-world ML problems with high-quality, validated training data. Conversely, it provides the marine science community access to algorithms that can help reduce annotation time and improve data quality. Traditionally, annotation tools that offer automation are wedded to a particular internal ML pipeline. This architecture offers unprecedented flexibility for ML integration.
Squidle+ also offers a semantic translation framework, which can be used to standardise the vocabularies by cross-walking between different label schemes. This facilitates data reuse, syntheses between projects and large-scale training of ML algorithms.
In this presentation we will showcase how the tools and features built into the Squidle+ platform facilitates automated annotation of marine imagery with transparency, quality control and unprecedented flexibility to accommodate multiple end user needs.