Generative AI models for video super-resolution

Description

The aim of the project is to develop a tool based on deep learning to improve the quality of video in two aspects: increase the resolution and synthesis of bokeh effect (keeping the subject sharp while blurring the background). For this purpose, generative techniques will be investigated (such as adversarial training, autoregressive models, variational autoencoders, diffusion models), which should be capable of generating sharp details in the focus plane that are temporally consistent with the motion in the video. The designed tool must comply with computational cost constraints to ensure its practical implementation.

The tasks will be the following:
* Design of data sets (possibly including synthetically generated data) for training and evaluation. Design, implementation and training of generative models for video.
* Tests on simple datasets to evaluate feasibility.
* Adaptation of the models to the complexities of real cases and optimization of the quality of the result.
* Reduction of the computational cost of the solution, affecting the quality of the results as little as possible.
* Design and implementation of a quantitative and qualitative evaluation protocol.

Qualifications

GENERAL REQUIREMENTS

a) Hold a Ph.D. degree

b) Not have been dismissed from the service of any public administration through disciplinary proceedings, nor be disqualified from public office by a final judgment. In the case of being a national of another state, not be disqualified or in an equivalent situation, nor have been subject to disciplinary or equivalent sanction that prevents, in your state, access to public employment under the same terms.

SPECIFIC REQUIREMENTS

a) Research experience in at least one of the following domains: computer vision,image and/or video processing, generative models for image generation, machine learning, deep learning.

b) Excellent programming skills and experience in developing deep learning pipelines using python libraries such as pytorch and tensorflow.

c) Capacity to work individually with autonomy and critical thinking, as well as in teams.

d) Good communication abilities.

e) Fluent level of English, both written as well as spoken.

These requirements must be met by the application deadline, must be reliably demonstrable at any point during the selection process, and must continue to be met on the contract signing date.

Application Instructions

Through Interfolio

Equal Employment Opportunity Statement

UPF promotes a diverse and inclusive environment and welcomes applicants regardless of age, disability, gender, nationality, race, religion or sexual orientation.