End of Master Project (TFM)

 

 

Join Us  

 

End of Master Project (TFM)  

 

2023

Evaluation and modelling of the user experience with immersive technologies

Recent years have witnessed many impressive technological and scientific advances relation to immersive technologies, such as Virtual Reality (VR), free-viewpoint video, 360° video, and Augmented Reality (AR). The availability of these technologies is paving the way to some extremely appealing new applications and services in different domains, such as entertainment, communications, social relations, healthcare, and industry. The users of these new technologies can explore and experience the contents in a more interactive and personalized way than previous technologies, intensifying their sensation of “being there”. These new perceptual dimensions and interaction behaviors provided by immersive technologies (e.g., exploration patterns, perceptual quality, immersiveness, simulator sickness, etc.), together with the new challenges concerning the whole processing chain (i.e., from acquisition to rendering), require an exhaustive study and understanding in order to satisfy the demands and expectations of the users. Thus, the evaluation of the user experience by subjective experiments with test participants and the development of models to predict and estimate it (e.g., based on machine learning techniques) is essential. In this sense, the Grupo de Tratamiento de Imágenes (GTI) has a long experience and is actively researching on subjective assessment tests and model development, and is looking for motivated students to work on these activities.

Contact person: Jesús Gutiérrez, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

*Other topics*

We can discuss about other possible topics (your own proposals are welcome!) related to evaluation/modelling of the user experience, eye tracking and visual attention models, virtual reality and image/video quality for health applications.

Contact person: Jesús Gutiérrez, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Deep learning for Semantic Segmentation and Depth Estimation

Deep learning has revolutionized Computer Vision with its remarkable performance. It can be applied to tasks such as Semantic Segmentation and Depth Estimation, which are closely related to immersive video technologies, volumetric capture systems and metaverse related applications.

The aim of this bachelor/master's thesis is to generate and/or annotate a dataset and to use it to train state-of-the-art deep learning models for semantic segmentation, depth estimation, and even combining both tasks.

We are looking for motivated students with a basic background in Deep Learning, Python, and Computer Vision.

Contact people:

Julián Cabrera: This email address is being protected from spambots. You need JavaScript enabled to view it.

Javier Usón: This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Video transmission for Cloud deployment of Deep Learning and Computer Vision applications

Deep Learning applications require heavy processing which must be performed by powerful machines. This is a huge problem for portable video setups as these machines are often too difficult to transport and cannot be deployed anywhere.

The aim of this bachelor/master's thesis is to develop a real-time video capture + transmission system that allows to separate the capture hardware from the heavy processing. This way, the heavy applications can be deployed in a cloud scenario and the complexity of the capture system is greatly reduced.

This project would require students who enjoy programming and have some background in video encoding. C++, Python and Linux will be extensively used.

Contact people:

Julián Cabrera: This email address is being protected from spambots. You need JavaScript enabled to view it.

Javier Usón: This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Real-time Video transmission for immersive applications

Immersive video technologies often require heavy processing, so they are restricted to powerful devices. This can be avoided by letting an external server perform the processing and transmitting the results and a regular 2D video. In this scenario, any device capable of playing video would be able to display the immersive experience.

The aim of this bachelor/master's thesis is to develop web and/or Unity applications that can communicate with an immersive video system so the experience can be controlled from different devices such as laptops, phones or head mounted displays.

We are looking for motivated students who enjoy programming and have some background in video encoding and transmission.

Contact people:

Julián Cabrera: This email address is being protected from spambots. You need JavaScript enabled to view it.

Javier Usón: This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Point Cloud Compression

Point clouds are used to represent volumetric visual data. A point cloud is a set of individual 3D points that contain some other attributes such as color, surface normal, etc.

The objective of this Bachelor/Master Thesis is to analyze the state of the art of Video-based point cloud compression (V-PCC) to identify the most suitable compression approaches for volumetric video. To assess the selected approaches a dataset of point clouds  captured with Microsoft Azure Kinect cameras will be also generated.

We are looking for motivated students with knowledge of video processing and encoding, C++ and Linux.

Julián Cabrera: This email address is being protected from spambots. You need JavaScript enabled to view it.

NeRFs for FVV

Neural Radiance Fields are a very promising approach to virtual view rendering using Neural Networks. They generate an implicit representation of a volumetric scene using a relatively small set of pictures of said scene, taken from different angles. This Bachelor/Masters Thesis involves researching the latest advancements in this field, apply them to a multi-camera setup and comparing its quality and speed to the ones obtained by a FVV system based on traditional rendering techniques.

We are looking for motivated students with knowledge of Deep Learning, Python and basic Computer Vision.

Julián Cabrera: This email address is being protected from spambots. You need JavaScript enabled to view it.

2022

Depth Image Generation and Compression for FVV

Free viewpoint video (FVV) is an application that allows the user to visualize a scene freely, choosing any arbitrary point of view they desire. This Bachelor/Master Thesis involves working with an implementation of FVV based around the use of depth images. It will involve researching ways of generating such depth images and then compress them in order to assess the impact these processes have on the final views rendered by the FVV system.

For Masters students, we propose a more in-depth dive into the depth image generation using Neural Networks for both mono and stereo imaging approaches.

This project would require students who enjoy programming and have some background in video encoding. C++, Python and Linux will be extensively used.

Contact person: Julián Cabrera, This email address is being protected from spambots. You need JavaScript enabled to view it.  

 

Development of a video encoder based on Intrinsically Non-Linear Receptive Fields

The development of efficient image and video compression algorithms is progressing thanks to the recent technological and machine-learning-based advances with the objective of delivering the best possible visual quality to the end users. In this sense, traditional image compression methods (e.g., JPEG, MPEG, etc.) use a fixed linear transforms for all bitrates, which, according to research studies, do not model the human visual system response as reliably as non-linear transforms. In particular, the Intrinsically Non-Linear Receptive Field (INRF), which describe the receptive field of a neuron, can be incorporated in image/video encoders using Deep Neural Networks (DNNs) to better reflect the human visual perception. Taking this into account, the Grupo de Tratamiento de Imágenes (GTI) is looking for motivated students to develop encoder for color videos from an existing grey-image encoder based on INRF and DNNs. 

Contact person: Jesús Gutiérrez, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

2021

Evaluation of the user experience with immersive technologies (VR, FVV, etc.)

Recent years have witnessed many impressive technological and scientific advances in relation to immersive technologies, such as Virtual Reality (VR), free-viewpoint video, 360° video, and Augmented Reality (AR). The availability of these technologies is paving the way to some extremely appealing new applications and services in different domains, such as entertainment, communications, social relations, healthcare, and industry. The users of these new technologies can explore and experience the contents in a more interactive and personalized way than previous technologies, intensifying their sensation of “being there”. These new perceptual dimensions and interaction behaviors provided by immersive technologies (e.g., exploration patterns, perceptual quality, immersiveness, simulator sickness, etc.), together with the new challenges concerning the whole processing chain (i.e., from acquisition to rendering), require an exhaustive study and understanding in order to satisfy the demands and expectations of the users. In this sense, the evaluation of the user experience by subjective experiments with test participants, and the development of models to predict and estimate it (e.g., based on machine learning techniques) is essential.

In this sense, the Grupo de Tratamiento de Imágenes (GTI) has a long experience and is actively researching on subjective assessment tests and model development, and is looking for motivated students to work on these activities within their TFGs or TFMs.

Contact person: Jesús Gutiérrez, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Visual attention modelling of people with autism spectrum disorder (ASD)

Visual attention (VA) allows to know the important regions of a scene for the observers, which is a crucial factor for many applications. Actually, modeling human VA has been a very active research topic during the last years, especially addressed by the vision science community, but also by the multimedia and computer vision communities. Recently, some studies showed that gaze features could be useful in the identification of mental states, cognitive processes and neuropathologies notably for people with Autism Spectrum Disorder (ASD). This has opened a new domain where VA modeling can be useful in different aspects, such as to help with the early stage detection of ASD, as well as to develop ad-hoc Computer–Human Interfaces (CHIs) that can be adapted for people with ASD. Therefore, this project aims to develop visual attention models (based on machine learning techniques) that can predict exploratory patterns of people with ASD and/or identify them from typically developed individuals from their visual attention patterns.

For this, we will work with existing public datasets and we will explore the possibility of building a new one in collaboration with the Hospital 12 Octubre.

Therefore, we are looking for motivated students to work on specific activities of this project within their TFGs or TFMs.

Contact person: Jesús Gutiérrez, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Design and implementation of a data augmentation system based on generative adversarial networks for exanthema images (Ref: Exanthema).

This Master Thesis addresses the problem of the lack of annotated data to train deep-learning-based algorithms in different areas of research. In this case we focus on a system to assist the diagnosis of skin problems in children that is being developed by the Grupo de Tratamiento de Imágenes. The objective of this Master Thesis is to evaluate  the use of generative adversarial networks to create new synthetic data samples as a data augmentation approach.

This Master Thesis will be done in collaboration with Hospital 12 de Octubre.

We are looking for students with experience with Python and Deep Learning Tools ( Keras, Tensorflow, PyTorch, …).

Contact person: Julián Cabrera, This email address is being protected from spambots. You need JavaScript enabled to view it.

  

Design and implementation of a depth estimation approach for Free Viewpoint Video environments using deep learning approaches (REF: DEPTH).

Free Viewpoint Video is an immersive multimedia system that provides the user the capability to move freely on a scene that is captured by a multiview set up of cameras. For the performance of the system, it is crucial to capture the geometric information of the scene (depth maps). This Master Thesis addresses the computation of depth maps using deep learning approaches. The scenario considered is a multiview set up in the framework of a specific Free Viewpoint Video (FVV) system, created by the Grupo de Tratamiento de Imágenes (GTI).

We are looking for students with Computer Vision background and experience with Python and Deep Learning Tools ( Keras, Tensorflow, PyTorch, …)

Contact person: Julián Cabrera, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

 

Design and implementation of a classification system for neuropathology microscopy images based on deep learning techniques (Ref: Neuro).

Microscopy images of the nervous system tissue are used to assist the diagnosis of disease. This Master Thesis addresses the design and development of a solution based on deep learning techniques to detect the presence of disease in these microscopy images. To that extend, an annotated database with images provided by the Hospital de Almería will be also created.

This Master Thesis will be done in collaboration with Hospital de Almería.

We are looking for students with experience with Python and Deep Learning Tools ( Keras, Tensorflow, PyTorch, …)

Contact person: Julián Cabrera, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

2020

Design and implementation of a diagnosis platform to assist Kawasaki disease identification

The Kawasaki disease is the most common heart condition affecting young children, usually under five years old, in developed countries. The disease is responsible for the damages of blood vessels all over the body and results in vasculitis, myocarditis and coronary dilation causing long term heart complications. Therefore, it is essential to be able to detect the disease at an early state.

One of the methods used to detect Kawasaki disease is by the analysis of the echocardiograms of the heart. In the Grupo de Tratamiento de Imágenes we have already developed classification and segmentation algorithms based on Deep Learning techniques for the echocardiograms. This Master Thesis will aim the integration and improvement of both algorithms within a common platform. Besides, this platform will also address the final diagnosis of the Kawasaki disease from echocardiograms and clinical data based also on Deep Learning techniques.

This Master Thesis will be done in collaboration with Hospital 12 de Octubre.

We are looking for students with experience with Python and Deep Learning Tools ( Keras, Tensorflow, PyTorch, …)

Contact person: Julián Cabrera, This email address is being protected from spambots. You need JavaScript enabled to view it.

 

Design and implementation of a breath-rate measurement solution based on computer vision and deep learning techniques

According to the last global estimate of pneumonia mortality, nearly one million children die from pneumonia worldwide every year, accounting for 16% of child deaths globally. Despite its prevalence, childhood pneumonia remains difficult to diagnose, particularly in those contexts where diagnostic imaging is unavailable and is based on subjective clinical signs and symptoms such as fast breathing and chest indrawing.

Accurate assessment of the respiratory rate is critical in Low Income Countries where other diagnostic tools, such as pulse oximetry or chest radiography, are not available. In this sense, this Master Thesis addresses the design and development of a solution for measuring the respiratory rate from a controlled video capture of the child using a smartphone. The approach will be based on computer vision and deep learning techniques.

This Master Thesis will be done in collaboration with Hospital 12 de Octubre.

We are looking for students with knowledge and experience with Python, Computer Vision techniques and SW tools, and Deep Learning Tools ( Keras, Tensorflow, PyTorch, …)

Contact person: Julián Cabrera, This email address is being protected from spambots. You need JavaScript enabled to view it.