Queen's University Belfast
School of Electronics, Electrical Engineering and Computer Science
- Location
- Belfast, United Kingdom
- Funding
- REF: EEECS27/GS2 - CSC-PHD
- Application deadline
- See university guidance
About the project
- Share Facebook Twitter Linkedin Email
- Postgraduate Research 2026
- PhD Opportunities
- Find a PhD Supervisor
Learning the World from Video: Compact 3D World Models for Spatial AI
- Learning the World from Video: Compact 3D World Models for Spatial AI: School of Electronics, Electrical Engineering and Computer Science | PHD
- Project details
- Subject information
- Course content
- Entry Requirements
- Fees and Funding
- Apply
Overview
When we watch someone move a chair across a room, we understand more than a sequence of images. We recognise the room’s layout, distinguish the moving chair from the moving camera, and expect the chair to remain present when it disappears behind another object. We can also imagine the scene from a different viewpoint. Developing comparable spatial understanding remains a major challenge for artificial intelligence: producing a convincing image does not necessarily mean having a consistent model of the world. This PhD will investigate how AI can learn compact, interpretable three-dimensional world models from ordinary video. The aim is to explain a scene using a camera trajectory and a small collection of simple, persistent 3D objects, then use that explanation to predict unseen views. Combining computer vision, self-supervised learning and differentiable graphics, the project will explore foundations for spatially aware AI, with potential applications in robotics, digital twins and immersive visualisation.
Research aim. The project will investigate whether learning to explain video through compact 3D scenes produces useful, robust spatial representations. A world model here means an explicit representation of scene layout, objects and their motion. The core task is to reconstruct and replay observed events; predicting future actions and large-scale video generation are possible extensions. The contribution will be new methods for learning object-level abstractions from video and controlled evidence about when these abstractions help. Research questions. (1) Can a model jointly learn camera motion and simple 3D scene structure without requiring manually annotated cameras or geometry? (2) How should it balance geometric simplicity with visual fidelity, and does abstraction improve robustness or data efficiency compared with dense and implicit representations? (3) Can persistent object representations separate camera motion from object motion, maintain identity through occlusion, and support consistent replay from new viewpoints? Learning by rendering. The student will develop a neural video encoder that estimates camera motion and represents a scene using a layout and a small set of geometric primitives, such as boxes and rounded shapes. A differentiable renderer will turn these predictions into images and visual features at held-out viewpoints. Comparing predictions with the corresponding video frames will provide the learning signal. The scene representation will be built only from context frames, with safeguards against target-image information leaking through camera estimation. A complexity penalty will favour concise explanations; coarse-to-fine image and feature losses will encourage geometry to capture structure while a limited appearance component handles texture. From static scenes to moving objects. Initial experiments will use procedurally generated rooms and curated static-scene videos, progressing to selected egocentric video clips. The model will then be extended with persistent object identities and time-varying poses, initially restricted to short clips with a few rigid moving objects. The student will test whether this representation preserves objects through occlusion and supports structural replay of an event from another camera position. Pretrained visual encoders and teacher-assisted initialisation may be used where appropriate, with their supervision and contribution documented separately from label-free training experiments. Evaluation. Experiments will use scene-disjoint training and test sets and suitable licensed benchmarks, for example ScanNet++ for geometry and Aria Digital Twin or Ego-Exo4D for motion and viewpoint evaluation where ground truth is available. Measures will include camera-pose error, layout and object recovery, identity retention through occlusion, held-out-view structural accuracy, and accuracy versus representation size. Frozen-encoder depth and pose probes will assess the learned representation’s transferability. Comparisons will include a dense reconstruction model, an implicit video representation and a reconstruction-then-primitive-fitting pipeline. Key ablations will match data, encoder initialisation and compute. Evaluation cameras will be independently supplied or aligned to reference trajectories; target appearance will not enter scene construction. Monocular geometry will be evaluated up to scale unless metric information is supplied. Programme and feasibility. Year 1 will establish baselines, a synthetic testbed and a static-scene prototype. Year 2 will investigate abstraction, robustness and transfer to real video. Year 3 will study a bounded moving-object setting, complete evaluation and demonstrations, and write the thesis. An additional six months, if supported by an EPSRC award, would enable broader validation. The core study will use pretrained components, modest model sizes and curated subsets, with compute requirements agreed with the supervisors. If joint learning proves unstable, staged camera estimation or an auxiliary primitive-prediction head will provide a tractable alternative. Foundation-scale training is not required for completion. Outcomes and training. Expected outputs include reproducible methods and benchmarks, research publications, and a demonstrator converting a short video into an inspectable 3D scene with alternative-view replay. The student will gain expertise in deep learning, 3D vision, graphics, experimental design and research software. Potential applications include simulator scene creation, spatial memory for AI assistants and interactive scene exploration. Responsible research. Work will use appropriately licensed datasets and respect restrictions on personal and egocentric video. Any new data collection will require relevant consent and ethical approval. Releases will respect data and model licences, and reporting will address failure cases, computational cost and potential misuse in surveillance or misleading media.
Funding Information
Funding Information To be eligible for consideration for a Home Studentship (covering tuition fees and maintenance stipend of approx. £21,805 per annum), a candidate must satisfy all the eligibility criteria based on nationality, residency and academic qualifications. To be classed as a Home student, candidates must meet the DfE/EPSRC residency requirements and the following criteria: • Be a UK National, or • Have settled status, or • Have pre-settled status, or • Have indefinite leave to remain or enter the UK. Irish nationals also qualify for Home student funding. Previous PhD study MAY make candidates ineligible to be considered for funding. Please note that other terms and conditions also apply. For further information about the residency and nationality requirements please see: DFE:https://www.economy-ni.gov.uk/sites/default/files/publications/economy/postgraduate-studentships-terms-conditions.pdf EPSRC: https://www.ukri.org/publications/terms-and-conditions-for-training-funding/ Please note that any available PhD studentships will be allocated on a competitive basis across a number of projects currently being advertised by the School. International Candidates: A small number of international awards will be available for allocation across the School. An international award is not guaranteed to be available for this project, and these awards will be highly competitive. International candidates are permitted to apply for a maximum of TWO projects across the entire school. Applications for more than two projects may be rejected due to a lack of relevance in research interests. Applicants must also upload complete documentation for their Bachelor’s degree and, where applicable, their Master’s degree. Failure to provide the required academic documentation may result in the application being marked as withdrawn. Academic Requirements: The minimum academic requirement for admission is normally an Upper Second Class Honours degree from a UK or ROI Higher Education provider in a relevant discipline, or an equivalent qualification acceptable to the University. Relevant disciplines include Computer Science, Artificial Intelligence, Electrical or Electronic Engineering, Mathematics, Physics, or a closely related subject. Applicants should demonstrate programming ability and a sound foundation in linear algebra, probability and mathematical reasoning. Experience with Python and a deep-learning framework, computer vision, computer graphics or robotics would be advantageous, but prior expertise in every area is not expected. A Master’s degree is not an additional project-specific requirement; admission and studentship eligibility remain subject to the University and funder criteria stated above.
Dr Guanxiong Sun
g.sun@qub.ac.uk
Research Profile
Full-time: 3 or 3.5 years
Computer Science overview
The School of Electronics, Electrical Engineering and Computer Science (EEECS) aims to enhance the way we use technology in communication, data science, computing systems, cyber security, power electronics, intelligent control, and many related areas. You’ll be part of a dynamic doctoral research environment and will study alongside students from over 40 countries world-wide. We supervise students undertaking research in key areas of computer science, including: - Artificial Intelligence - Cybersecurity - Computing Systems - Power Electronics - Robotics - Sensor-based Systems - Wireless Communications Within the School we have a number of specialist research centres. As part of a lively community of over 100 full-time and part-time research students you’ll have the opportunity to develop your research potential in a vibrant research community that prioritises the cross-fertilisation of ideas and innovation in the advancement of knowledge. Many PhD studentships attract scholarships and top-up supplements. PhD programmes provide our students with the opportunity to acquire an extensive training in research techniques.
Research students are encouraged to play a full and active role in relation to the wide range of research activities undertaken within the School and there are many resources available including:
- Queen’s researchers have strong links with the local industry, which boasts a rich mix of local startups and multi-nationals. Belfast is the second fastest growing region in the UK in terms of Knowledge Economy activity (Northern Ireland Economy Report, 2018).
- CSIT brings together research specialists in complementary fields such as data security, network security systems, wireless-enabled security systems, intelligent surveillance systems; and serves as the national point of reference for knowledge transfer in these areas.
- The state-of-the-art £14m Computer Science Building and the Institute of Electronics, Communications and Information Technology offer bespoke research environments. The Institute of Electronics, Communications and Information Technology (ECIT), with state-of-the-art technology, offers a bespoke research environment.
- You will be working under the supervision of leading international academic experts.
- A wide range of personal development and specialist training courses offered through the Personal Development Programme
- Access to the Queen's University Postgraduate Researcher Development Programme
- Office accommodation with access to computing facilities and support to attend conferences for full-time PhD students
Course content
Associated Research Research within the School is organised into research themes combining strengths by working together on major projects, in many cases in collaboration with key technology companies. ECIT brings together internationally recognised research groups specialising in key areas of advanced digital and communications technology.
PhD Opportunities PhD Opportunities are available in a wide range of computer science subjects, aligned to the specific expertise of our PhD supervisors.
Research Impact Queen’s is a leader in commercial impact and one of the five highest performing universities in the UK for intellectual property commercialisation. We have created over 80 spin-out companies. Three of these - Kainos, Andor Technology and Fusion Antibodies - have been publicly listed on the London Stock Exchange.
Research Projects Queen’s has strong collaborative links with industry in Northern Ireland, and internationally. It has a strong funding track record with EPSRC and the EC H2020 programme.
Research Success The research profile produced by the 2014 UK Research Excellence Framework (REF) graded 80 per cent of our research activity as 'world-leading' or 'internationally excellent', confirming the School's reputation as an internationally-leading department.
Introduction For further information on career opportunities at PhD level please contact the Faculty of Engineering and Physical Sciences Student Recruitment Team on askEPS@qub.ac.uk. Our advisors - in consultation with the School - will be happy to provide further information on your research area, possible career prospects and your research application.
Full time PhD students will have access to a shared office space and access to a desk with personal computer and internet access.
- Assessment processes for the Research Degree differ from taught degrees. Students will be expected to present write up their work at regular intervals to their supervisor who will provide written and oral feedback; a formal assessment process takes place annually. This Annual Progress Review requires students to present their work in writing and orally to a panel of academics from within the School. Successful completion of this process will allow students to register for the next academic year. The final assessment of the doctoral degree is both oral and written. Students will submit their thesis to an internal and external examining team who will review the written thesis before inviting the student to orally defend their work at a Viva Voce.
- Supervisors will offer feedback on the research work at regular intervals throughout the period of registration on the degree.
Entrance requirements
Graduate The minimum academic requirement for admission to a research degree programme is normally an Upper Second Class Honours degree from a UK or ROI HE provider, or an equivalent qualification acceptable to the University. Further information can be obtained by contacting the School.
For information on international qualification equivalents, please check the specific information for your country .
Evidence of an IELTS score of 6.0, with not less than 5.5 in any component or equivalent qualification acceptable to the University is required (taken within the last 2 years).
International students wishing to apply to Queen's University Belfast (and for whom English is not their first language), must be able to demonstrate their proficiency in English in order to benefit fully from their course of study or research. Non-EEA nationals must also satisfy UK Visas and Immigration (UKVI) immigration requirements for English language for visa purposes.
For more information on English Language requirements for EEA and non-EEA nationals see: www.qub.ac.uk/EnglishLanguageReqs .
If you need to improve your English language skills before you enter this degree programme, INTO Queen's University Belfast offers a range of English language courses. These intensive and flexible courses are designed to improve your English ability for admission to this degree.
Tuition Fees
Northern Ireland (NI) 1 | TBC
Republic of Ireland (ROI) 2 | TBC
England, Scotland or Wales (GB) 1 | TBC
EU Other 3 | £28,900
International | £28,900
1 EU citizens in the EU Settlement Scheme, with settled or pre-settled status, are expected to be charged the NI or GB tuition fee based on where they are ordinarily resident, however this is provisional and subject to the publication of the Northern Ireland Assembly Student Fees Regulations. Students who are ROI nationals resident in GB are expected to be charged the GB fee, however this is provisional and subject to the publication of the Northern Ireland Assembly student fees Regulations.
2 It is expected that EU students who are ROI nationals resident in ROI will be eligible for NI tuition fees. The tuition fee set out above is provisional and subject to the publication of the Northern Ireland Assembly student fees Regulations.
3 EU Other students (excludes Republic of Ireland nationals living in GB, NI or ROI) are charged tuition fees in line with international fees.
All tuition fees quoted relate to a single year of study unless stated otherwise. All fees will be subject to an annual inflationary increase, unless explicitly stated otherwise.
More information on postgraduate tuition fees .
Computer Science costs
There are no specific additional course costs associated with this programme.
Additional course costs
Depending on the programme of study, there may also be other extra costs which are not covered by tuition fees, which students will need to consider when planning their studies . Students can borrow books and access online learning resources from any Queen's library. If students wish to purchase recommended texts, rather than borrow them from the University Library, prices per text can range from £30 to £100. Students should also budget between £30 to £100 per year for photocopying, memory sticks and printing charges. Students may wish to consider purchasing an electronic device; costs will vary depending on the specification of the model chosen. There are also additional charges for graduation ceremonies, and library fines. In undertaking a research project students may incur costs associated with transport and/or materials, and there will also be additional costs for printing and binding the thesis. There may also be individually tailored research project expenses and students should consult directly with the School for further information.
Some research programmes incur an additional annual charge on top of the tuition fees, often referred to as a bench fee. Bench fees are charged when a programme (or a specific project) incurs extra costs such as those involved with specialist laboratory or field work. If you are required to pay bench fees they will be detailed on your offer letter. If you have any questions about Bench Fees these should be raised with your School at the application stage. Please note that, if you are being funded you will need to ensure your sponsor is aware of and has agreed to fund these additional costs before accepting your place.
How do I fund my study?
Find PhD opportunities and funded studentships by subject area.
We offer numerous opportunities for funded doctoral study in a world-class research environment. Our centres and partnerships, aim to seek out and nurture outstanding postgraduate research students, and provide targeted training and skills development.
The Government offers doctoral loans of up to £26,445 for PhDs and equivalent postgraduate research programmes for English- or Welsh-resident UK and EU students.
Information on Postgraduate Research scholarships for international students .
The Funding & Scholarship Finder helps prospective and current students find funding to help cover costs towards a whole range of study related expenses.
How to Apply
Apply using our online Postgraduate Applications Portal and follow the step-by-step instructions on how to apply .
If you're interested in a particular project, we suggest you contact the relevant academic before you apply, to introduce yourself and ask questions.
To find a potential supervisor aligned with your area of interest, or if you are unsure of who to contact, look through the staff profiles linked here .
You might be asked to provide a short outline of your proposal to help us identify potential supervisors.
Download Postgraduate Prospectus