The platform is designed to support scientists and investigators in academic institutes, environmental agencies, public bodies as well as environmental consultancy companies. The services for management and analysis of marine data are open-source and can be used by anyone. If you want to upload and analyse your footage on SUBSIM, please contact us by email matthias[dot]obst[at]marine.gu.se, and we will help you to get started.

SUBSIM currently consists of five modules detailed  below. Each module represents a basic function in a image analysis project using machine-learning. These modules can be executed independently, which allows users to enter the analytical cycle at any point and only use the functions needed. 

Module 1: Project setup

This module helps you to  set up an machine learning project, prepare the metadata in the right format to ensure provenance in the data flow and make the data outputs FAIR and AI-ready. 

Module 2: Model training and evaluation 

This module takes labelled images ad input to train a computer vision model (for object detection or segmentation). You can then fine-tune, evaluate, and further optimise the model. 

Module 3: Model inference

This module uses a trained object detection or segmentation model to make predictions on large volumes of unseen videos or images. 

Module 4: Post-processing

This modules analyzes your inference results with summary statistics and vizualisations. 

Module 5. Publishing model and data products

This module guides you through the process of publishing your final model and data products to public archives and biodiversity data platforms.

When you use SUBSIM, you start with organising your media and metadata and then upload these to a workspace in a high-performance comuting center, see (a) in the below figure. Thereafter you can label images and movies, using third-party tools (b). Once you have created training data, you can train machine-learning models to perform e.g., object detection or image segmentation (c). Once your model is well trained and evaluated, it can be published (e) and deployed to run inference at scale on your media in a high-performance computing environment (d). Finally, there is also the option to publish the machine-based species observations to the Global Biodiversity Information Facility, GBIF (e). 

Here you find the repo for the development of the Swedish Biodiversity Data Infrastructure (SBDI), to which SUBSIM is connected and delivers data to.