—————————————————————————————————
Program Name: Telecommunications
Department: Department of Electrical Engineering
Degree Name* : PhD
Study Level* : Doctorate
Course Intensity*: Part-Time
Study Mode*: On Campus
MBA Program Type: Part-Time MBA
Program Details
Broad Subject Area*: Telecommunications
Main Subject*: Telecommunications Systems
Custom Subject: Electrical networks, Electrical control, Electrical machines.
Specialization: ……………………………………………………
Program Description:The research will address the development of deep solutions to face analysis that are based on Transformers. These new paradigms have not been used in the domain of face analysis. Obviously, a key issue in Transformers is how one can generate the input sequence for a given face image. This can be a main challenge in applying the Transformers to face images. To this end, several strategies will be proposed and tested. Among which the technique that deploys Multi-scale Multi-level for the input face will be explored. Another objective of the doctoral training is to develop or explore the use of deep Transformers in the domain of facial video analysis. Thus, instead of developing a way by which the individual image descriptors or decisions can be fused in order to characterize the face dynamics, a direct solution is thought. The doctoral studies will evaluate several strategies by which Transformers can be trained and tested on the video sequence. For instance a first strategy is to consider the video frames as a individual patches that are stamped in the time line. Other strategies can consider the patches in the 3D volume of the video.
The developed algorithms and tools will be applied to the recent problems of age estimation.
The research will address the development of deep solutions to face analysis that are based on Transformers. Recent applications in computer vision like dynamic face expression recognition, driver fatigue detection, pain assessment, age estimation, kinship verification, face beauty prediction, and personality traits estimation relied on the analysis of human facial images. Deployed techniques can rely either on the analysis of a single snapshot of the face or on the analysis of a sequence of snapshots. Human facial image analysis is among the most challenging areas in computer vision research. This is because the task in hand is to design and extract a discriminant descriptor that takes into account the spatiotemporal aspect of face images. Unlike natural scenes (oceans, tree leaves, fountains, streets), analyzing dynamic face textures can be more difficult since faces undergoes rigid and non-rigid motions and since faces can be partially occluded.
Use of Transformers in Vision tasks: The use of Transformers in solving image-based tasks is relatively new . “AN IMAGE IS WORTH 16X16 WORDS: TRANSFORMERS FOR IMAGE Recognition at Scale”, a new paradigm for deep learning is used in some works. This is given by the vision Transformers. These neural networks do not use convolution and instead they use self-attention mechanisms that exploit the sequence of patches where it is explored the direct application of Transformers to image recognition. The image can be interpreted as a sequence of patches and process it by a standard Transformer encoder as used in Natural Language Processing NLP. Vision Transformer (ViT) provides good results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train. For a given amount of compute, Vision Transformers yield better performance than the equivalent Convolutional Neaural Networks (CNNs). This doctoral training aims to develop and explore the new deep learning paradigm (Transformers) in the domain of face image analysis. The resulting tools can be used by some emerging techniques such as age estimation, kinship verification, face beauty assessment, and pain assessment.
Get more details (the responsible email)*: [email protected]
Students per Class: 12
Average age (in years): 30
Average years of work experience at managerial level:Administration staff
Percentage of international students: 0%
Percentage of women: 30%
Average salary after graduation: 250 USD
Percent employment after graduation: 70%
Average work experience (in years): Just in the current position
Number of nationalities in current cohort : 1