PhD Studentship: Synthetic-to-Real Learning for Fish Segmentation, Classification and Shape Analysis

Primary supervisor - Prof Michal Mackiewicz 

Accurate visual assessment of fish on board fishing vessels, in processing facilities and in underwater environments is vital for monitoring migrating fish, fishing activities, stock assessment, and product quality. Yet, automated processing of fish in real-world is highly challenging: fish move unpredictably, appear under variable lighting and water conditions, and are often partially occluded by other fish, equipment, or people. These factors make the development of automated fish segmentation, classification and measurement difficult, while large, annotated datasets required for training robust AI models are costly to obtain. 

This PhD project will develop new synthetic-to-real learning methods for robust fish segmentation, classification, and shape analysis with minimal manual labelling. The successful candidate will first create a controllable 3D rendering pipeline (e.g. using Blender) to generate large, photo-realistic synthetic datasets of fish under diverse lighting, orientation, and occlusion scenarios, providing precise ground truth including segmentation masks, 2D and 3D shape, and species labels. 

Building on these datasets, the research will investigate self- and semi-supervised learning approaches to pretrain models that capture fish morphology and spatial structure, even when parts of the body are hidden. The project will also explore domain-adaptation strategies to bridge the gap between synthetic and real on-board and underwater imagery. The goal is a domain-adaptive framework that generalises across environments and species, capable of accurately segmenting occluded fish, classifying species, and extracting 2D/3D shape metrics such as length and weight from real video streams. 

The project offers collaboration with computer vision researchers, marine scientists, and industry partners across the current closely aligned Horizon Europe EVERYFISH project and past SMARTFISH and AVIMS projects and will leverage real-world fish datasets collected through these projects. 

Applicants should have experience in computer vision and deep learning. Skills in 3D graphics are desirable but not essential. 

Entry requirements

The standard minimum entry requirement is 2:1 in Computer Science/Physics/Maths or other numerate discipline. 

Mode of study

Full-time

Start date

1 October 2026

Funding

This PhD project is in a competition for a Faculty of Science funded studentship. Funding is available to UK applicants and comprises ‘home’ tuition fees and an annual stipend for 3 years.

Closing Date

10/12/2025

To apply for this role, please click on the 'Apply' button above.

Advert information