Student Projects

LTS5 OPEN SEMESTER AND MASTER PROJECTS – Fall 2020

SEMESTER PROJECT PROPOSALS

1. Wrinkles Modeling

Human face has always been of particular interest in the computer graphics community. Because of its complexity, modeling lifelike synthetic objects is challenging. A variety of approaches have been proposed such as statistical models (i.e. principal component analysis models) or blendshapes models.  Moreover, tackling the variation in terms of population (*i.e. identity*) and expression at the same time in a generic 3D model increases the difficulty.

With conventional modelling technics, the detailed facial attributes such as the wrinkles are lost in the process. Moreover these mid-frequencies informations are important for photo-realistic expressions generation. Therefore an explicit model can be jointly used to augment the original face model to recover them.

The goals of this project are in two folds. Develop a solution to recover the details (*i.e face wrinkles*) that are not present with standard statistical linear models. Finally, build an explicit model from the the previously extracted data and combine everything together for photo-realistic face generation (see figure below).

Requirements:  The project will be implemented in Python / C++ so good knowledge is required. Previous experience in one or several of the following topics would be a plus: image processing, computer vision or machine learning.

Assistant: Christophe Ecabert ([email protected])

Supervisor: Prof. Jean-Philippe Thiran

2. Apizoom – deep learning to quantify the Verroa parasite in honey bee hive images

Varroa mites are recognized as the biggest pest to honey bees worldwide, and are believed to be the single largest contributing factor in the modern-day decline of honey bees due to their ability to transmit diseases, resulting in death or severe deformity of the pupae.

Verroa on honey bees.

Detecting and quantifying the presence of Verroa in a beehive is therefore crucial to treat the infection appropriately and as early as possible, and image analysis appears very useful in this problem.

In this project, we propose to develop an image analysis to detect and count Verroa cadavers who felt down on a plate below the beehive, as a non-intrusive way to quantify the presence of the Verroa. High definition images will be capture and Deep Learning techniques will be investigated here, to design and train a Convolutional Neural Network (CNN) to detect the Verroa and distinguish it from other wastes. See the following video for more details on the project (in French).

Depending on the evolution of the project, several steps will be investigated:

  • Development of the Deep Learning method
  • Training on a collection of annotated images
  • Test and improve

And possibly:

  • Study integration in mobile phones
  • Study the development of a web-based analysis system.

This project is jointly proposed with the company Apizoom (Fribourg, Switzerland).

Responsible: Prof. J.-Ph. Thiran

3. CleanCityIndex – A Deep Learning based system to localize and classify wastes on the streets

A review of major European cities places “urban cleanliness” as a top priority for the authorities, as it directly impacts the concern and satisfaction of their citizens and the attractiveness of their economy and tourism. Littering quantification is an important step toward improving urban cleanliness. When human interpretation is too cumbersome or in some cases impossible, an objective index of cleanliness could reduce the littering by awareness actions.

The goal of this project is to propose a fully automated computer vision application for littering quantification, based on images taken from the streets and sidewalks. We employ a deep learning based framework to localize and classify different types of wastes.

In this project, the student will be involved into study and develop new detection algorithms and investigates deep learning based object tracking techniques for this specific case. The student is expected to be familiar with Python and Tensorflow.

Assistant: Saeed Rad ([email protected])
Supervisor: Prof. Jean-Philippe Thiran

4. Abnormality Detection for X-Ray Radiographic Images

The X-ray radiograph (XR) such as chest X-ray is frequently acquired for detecting and diagnosing pathologies and it is widely used in cancer prevention and treatment planning. However, the radiologist reporting of every image is a tedious and time-consuming task. Having said that, abnormal cases are usually much rarer than normal cases . In addition, it is very time-consuming to collect a reasonable amount for abnormal cases, while on the other hand, normal XRs are much easier to obtain. Therefore, fully supervised deep neural networks’ (DNNs) effectiveness is reduced. Consequently, an automatic system of XR abnormality detection would be advantageous, allowing reducing time for annotation and paying more attention on pathology analysis of abnormal XRs. The goal of this project is to develop a computer-aided diagnosis (CAD) system that can automatically identify abnormal XRs by learning only from normal ones during training phase.

In this project, the student will be involved in study and develop DNN-based one-class classifier for abnormal X-ray detection. The student is expected to be familiar with Python and Pytorch.

Assistant: Dr Behzad Bozorgtabar ([email protected])

Supervisor: Prof. Jean-Philippe Thiran

5. Image Modality Conversion

 Recently, interest in using image modality conversion for radiotherapy applications have grown rapidly. For example, for replacing Computed tomography (CT) with magnetic resonance imaging (MRI) for diagnostic and therapeutic purposes. This is mainly because MRI delivers superior contrast of soft tissue compared with the CT scans and is free of ionizing radiation. Therefore, modality conversion is desirable for many applications such as CT-based radiotherapy. Recent studies have shown remarkable success in image-to-image translation for modality conversion. This task has experienced significant improve- ments following the introduction of generative adversarial networks (GANs).

In this project, the student will be involved in study and develop a DNN-based image synthesis approach for wide range of modality conversion with the focus on radiotherapy applications. The student is expected to be familiar with Python and Pytorch.

Assistant: Dr Behzad Bozorgtabar ([email protected])

Supervisor: Prof. Jean-Philippe Thiran

6. Deep learning for speed of sound ultrasound imaging

The last years have seen the development of a new modality for ultrasound imaging, namely speed of sound imaging in pulse-echo mode. Its basic principle is to compute locally the speed of sound inside the soft tissue of the body based solely on the wave reflected back by those tissues. Such a modality allows the extraction of meaningful informations for medical diagnoses. Being able to distinguish malignant and benign liver tumors is a typical example of application of such a technique.

The basic principle of most of the state-of-the art methods is to compare  the ultrasound images reconstructed trough delay-and-sum under different conditions (typically the angle of the plane-wave we send) and deduce from the local spatial differences between the image the map of speed of sound. This comparison is usually performed with traditional image processing technique suffering from several drawbacks.

Our idea is to replace the traditional image processing techniques by a neural network specifically trained for this task, a CNN inspired by what is done in the image processing community. The student will be involved in the development and the training of such a network. Depending on the availability of the data, different type of techniques requiring less data can be investigated as well.

Requirements: Python (PyTorch / Tensorflow), a good knowledge in signal processing and machine learning is preferable.

Supervisor: Jean-Philippe Thiran

Assistant: Samuel Beuret ([email protected]

7. Dynamic image analysis for Cervical Cancer Detection: methods and Android demonstrator – analyse d’images dynamiques pour la détction du cancer du col de l’uterus: méthodes et démonstreteur Android

Cervical cancer is a major concern in public health, both in developed and developing countries. Especially in this later context, the availability of well-trained experts is limited, and computer-aided diagnosis is clearly needed for large-scale screening.

This project will focus on the analysis of dynamic image sequences (videos) of the cervix under a contrast agent: the visual inspection with acetic acid (VIA) is known as one of the reference methods to detect cervical cancer. However the human eye has limited capabilities in assessing the time evolution of the appearance of the cervix after administration of the contrast agent. Therefore, in this phase of the project, the intensity curves of each pixel of the video will be analyzed by a machine-learning algorithm. A synthetic image displaying the results will be produced as a help for the diagnosis. Moreover, as a simple and portable tool is needed, especially in developing countries, an Android application will be developed to perform this analysis and display the results on mobile platforms.

This project will be realized in close collaboration with the Department of Gynecology of the Geneva University Hospital.

Requirements: C++, image processing, skills in development for Android would be ideal.

Supervisor: Prof. Jean-Philippe Thiran

8. Computer vision for the EPFL Racing Team

EPFL Racing team is a student team that competes in the Formula Student competition, the largest student competition for engineers in the world. We design and build electric race cars and compete against hundreds of other universities during summer, all across Europe.

Teams are scored on different ‘events’ such as car acceleration, endurance, agility and are driven by pilots that are students, members of the team. Performance wise, to give an example the 0 – 100 km/h is realized in approximately 2.8 sec!

Starting in summer 2021 we plan on having an autonomous car, able to visualize the track and to find the perfect line to get the best lap time! To do so, we are looking to start projects on the topics of computer vision.

The aim of the projects would be to find and implement the most performant and robust detection algorithms, knowing that the track is bounded by yellow and blue cones (cf. picture of the car of ETH in 2017). The car must be as quick as possible around the track while hitting as few cones as possible, otherwise we get a time penalty. Also, it will be crucial to consider the hardware implementation on the 2021 car: what sensors, where to put them and how much power is required.

Supervisor: Prof. Jean-Philippe Thiran

 

MASTER PROJECT PROPOSALS 

MEDICAL IMAGING PROJECTS

1. Abnormality Detection for X-Ray Radiographic Images

The X-ray radiograph (XR) such as chest X-ray is frequently acquired for detecting and diagnosing pathologies and it is widely used in cancer prevention and treatment planning. However, the radiologist reporting of every image is a tedious and time-consuming task. Having said that, abnormal cases are usually much rarer than normal cases . In addition, it is very time-consuming to collect a reasonable amount for abnormal cases, while on the other hand, normal XRs are much easier to obtain. Therefore, fully supervised deep neural networks’ (DNNs) effectiveness is reduced. Consequently, an automatic system of XR abnormality detection would be advantageous, allowing reducing time for annotation and paying more attention on pathology analysis of abnormal XRs. The goal of this project is to develop a computer-aided diagnosis (CAD) system that can automatically identify abnormal XRs by learning only from normal ones during training phase.

In this project, the student will be involved in study and develop DNN-based one-class classifier for abnormal X-ray detection. The student is expected to be familiar with Python and Pytorch.

Assistant: Dr Behzad Bozorgtabar ([email protected])

Supervisor: Prof. Jean-Philippe Thiran

2. Image Modality Conversion

 Recently, interest in using image modality conversion for radiotherapy applications have grown rapidly. For example, for replacing Computed tomography (CT) with magnetic resonance imaging (MRI) for diagnostic and therapeutic purposes. This is mainly because MRI delivers superior contrast of soft tissue compared with the CT scans and is free of ionizing radiation. Therefore, modality conversion is desirable for many applications such as CT-based radiotherapy. Recent studies have shown remarkable success in image-to-image translation for modality conversion. This task has experienced significant improve- ments following the introduction of generative adversarial networks (GANs).

In this project, the student will be involved in study and develop a DNN-based image synthesis approach for wide range of modality conversion with the focus on radiotherapy applications. The student is expected to be familiar with Python and Pytorch.

Assistant: Dr Behzad Bozorgtabar ([email protected])

Supervisor: Prof. Jean-Philippe Thiran

3. Deep learning for speed of sound ultrasound imaging

The last years have seen the development of a new modality for ultrasound imaging, namely speed of sound imaging in pulse-echo mode. Its basic principle is to compute locally the speed of sound inside the soft tissue of the body based solely on the wave reflected back by those tissues. Such a modality allows the extraction of meaningful informations for medical diagnoses. Being able to distinguish malignant and benign liver tumors is a typical example of application of such a technique.

The basic principle of most of the state-of-the art methods is to compare  the ultrasound images reconstructed trough delay-and-sum under different conditions (typically the angle of the plane-wave we send) and deduce from the local spatial differences between the image the map of speed of sound. This comparison is usually performed with traditional image processing technique suffering from several drawbacks.

Our idea is to replace the traditional image processing techniques by a neural network specifically trained for this task, a CNN inspired by what is done in the image processing community. The student will be involved in the development and the training of such a network. Depending on the availability of the data, different type of techniques requiring less data can be investigated as well.

Requirements: Python (PyTorch / Tensorflow), a good knowledge in signal processing and machine learning is preferable.

Supervisor: Jean-Philippe Thiran

Assistant: Samuel Beuret ([email protected])

4. Dynamic image analysis for Cervical Cancer Detection: methods and Android demonstrator – analyse d’images dynamiques pour la détction du cancer du col de l’uterus: méthodes et démonstreteur Android

Cervical cancer is a major concern in public health, both in developed and developing countries. Especially in this later context, the availability of well-trained experts is limited, and computer-aided diagnosis is clearly needed for large-scale screening.

This project will focus on the analysis of dynamic image sequences (videos) of the cervix under a contrast agent: the visual inspection with acetic acid (VIA) is known as one of the reference methods to detect cervical cancer. However the human eye has limited capabilities in assessing the time evolution of the appearance of the cervix after administration of the contrast agent. Therefore, in this phase of the project, the intensity curves of each pixel of the video will be analyzed by a machine-learning algorithm. A synthetic image displaying the results will be produced as a help for the diagnosis. Moreover, as a simple and portable tool is needed, especially in developing countries, an Android application will be developed to perform this analysis and display the results on mobile platforms.

This project will be realized in close collaboration with the Department of Gynecology of the Geneva University Hospital.

Requirements: C++, image processing, skills in development for Android would be ideal.

Supervisor: Prof. Jean-Philippe Thiran

5. Self-supervised learning for computational pathology and survival prediction.

Computational pathology is a state-of-the-art technology that relies on histopathological images to predict a large variety of components such as tumor regions, clinical scores, or patient overall survival. Standard supervised machine learning approaches for tissue classifications and representations are limited by the amount of labeled data which are often rare in the medical field. Moreover, they often struggle to handle, in an efficient way, the large amount of data generated by the scanners.

Self-supervised learning is a branch of unsupervised learning that takes advantage of the structure of the data itself to learn meaningful features representations and, therefore, does not rely on labeled data. In addition, as the supervised aspect of the learning process directly comes from the data themselves, the models can benefit from the large quantity of generated data.

In this project, the student will use state-of-the-art self-supervised models to learn discriminative features for several medical tasks such as cancer region detection and patient survival prediction. The student is expected to be familiar with Python and Pytorch.

Assistant: Christian Abbet ([email protected])
Supervisor: Prof. Jean-Philippe Thiran

 

6. Assessing selective damage in cerebellar pathways in multiple sclerosis patients with optic myelitis.
Multiple Sclerosis (MS) is a chronic disease of the central nervous system characterized by brain inflammation, demyelination and axonal loss. A number of acute clinical syndromes in MS are known to occur due to inflammatory-demyelinating lesions in the brain stem and cerebellum, while chronic disabling symptoms (e.g. ataxia, oculomotor disturbances) are known to result from chronic damage in the posterior fossa. In this region, preliminary studies point to a selective damage of specific white matter tracts. However, the identification of these tracts has been elusive, with relatively few in vivo human studies to date.

In this study we propose to use state-of-the-art tractography methods together with microstructural parameters derived from diffusion and quantitative MRI acquisitions to delineate brainstem fiber pathways and define the most vulnerable tracts following MS of this key brainstem circuitry.
The student will work in close collaboration with Centre Hospitalier National d’Ophtalmologie des Quinze-Vingts in Paris and the Siemens Advanced Clinical Imaging Technology group at the EPFL Innovation Park. The project will be implemented in Python. Previous knowledge in medical imaging analysis and medical imaging software (FSL, Freesurfer, dipy, nipy) is a plus.

Supervisor: Prof. Jean-Philippe Thiran, Dr. Elda Fischi-Gomez ([email protected])

 

7.MRI-based automatic pathology characterization in the brain (master project with Siemens Healthcare, Innovation Square Lausanne)

Magnetic Resonance Imaging (MRI) is a well‐established medical imaging modality exhibiting excellent soft tissue contrast. typically weighted toward one or more physical properties that discriminate different tissues. More advanced quantitative MRI techniques allow moving from such relative contrast information to a single, absolute measure of one or more separate tissue properties, hence providing the means for tissue characterization and the potential to gain further insight into microstructural changes caused by diseases. The brevity and robustness of recently developed fast MR relaxometry methods allow for establishing normative atlases of quantitative parameters in healthy tissues which are sensitive to subtle tissue alterations on a single-subject basis – bearing great potential for clinical decision support [Piredda et al., MRM, 2020]. However, the establishment of voxel-wise normative atlases is hindered by the large anatomic inter‐subject variation of the cortices that often leads to their imperfect spatial normalization. In fact, this slightly flawed matching of cortices across subjects often results in a mixed distribution of white and gray matter, rendering the interpretation of results in cortices problematic. Objective of the proposed thesis is to explored advanced image processing pipeline, such as those offer by FreeSurfer or ANTs, to extend the current methodology to cortices. The established framework will be ultimately validated in single-subject analysis of patient’s data.

Required skills: Background in physics, computer science, electrical engineering, biomedical engineering, or similar. Previous programming experience in Python and MATLAB.

Company Information: Siemens Healthcare AG ACIT – EPFL QI-E, 1015 Lausanne, Switzerland.

Contact Information: Dr Tobias Kober, email: [email protected]

 

COMPUTER VISION PROJECTS

​​​​8. Wrinkles Modeling

Human face has always been of particular interest in the computer graphics community. Because of its complexity, modeling lifelike synthetic objects is challenging. A variety of approaches have been proposed such as statistical models (i.e. principal component analysis models) or blendshapes models.  Moreover, tackling the variation in terms of population (*i.e. identity*) and expression at the same time in a generic 3D model increases the difficulty.

With conventional modelling technics, the detailed facial attributes such as the wrinkles are lost in the process. Moreover these mid-frequencies informations are important for photo-realistic expressions generation. Therefore an explicit model can be jointly used to augment the original face model to recover them.

The goals of this project are in two folds. Develop a solution to recover the details (*i.e face wrinkles*) that are not present with standard statistical linear models. Finally, build an explicit model from the the previously extracted data and combine everything together for photo-realistic face generation (see figure below).

Requirements:  The project will be implemented in Python / C++ so good knowledge is required. Previous experience in one or several of the following topics would be a plus: image processing, computer vision or machine learning.

Assistant: Christophe Ecabert ([email protected])

Supervisor: Prof. Jean-Philippe Thiran

9. Apizoom – deep learning to quantify the Verroa parasite in honey bee hive images (can be considered as a Master project in industry)

Varroa mites are recognized as the biggest pest to honey bees worldwide, and are believed to be the single largest contributing factor in the modern-day decline of honey bees due to their ability to transmit diseases, resulting in death or severe deformity of the pupae.

Verroa on honey bees.

Detecting and quantifying the presence of Verroa in a beehive is therefore crucial to treat the infection appropriately and as early as possible, and image analysis appears very useful in this problem.

In this project, we propose to develop an image analysis to detect and count Verroa cadavers who felt down on a plate below the beehive, as a non-intrusive way to quantify the presence of the Verroa. High definition images will be capture and Deep Learning techniques will be investigated here, to design and train a Convolutional Neural Network (CNN) to detect the Verroa and distinguish it from other wastes. See the following video for more details on the project (in French).

Depending on the evolution of the project, several steps will be investigated:

  • Development of the Deep Learning method
  • Training on a collection of annotated images
  • Test and improve

And possibly:

  • Study integration in mobile phones
  • Study the development of a web-based analysis system.

This project is jointly proposed with the company Apizoom (Fribourg, Switzerland).

Responsible: Prof. J.-Ph. Thiran

10. CleanCityIndex – A Deep Learning based system to localize and classify wastes on the streets

A review of major European cities places “urban cleanliness” as a top priority for the authorities, as it directly impacts the concern and satisfaction of their citizens and the attractiveness of their economy and tourism. Littering quantification is an important step toward improving urban cleanliness. When human interpretation is too cumbersome or in some cases impossible, an objective index of cleanliness could reduce the littering by awareness actions.

The goal of this project is to propose a fully automated computer vision application for littering quantification, based on images taken from the streets and sidewalks. We employ a deep learning based framework to localize and classify different types of wastes.

In this project, the student will be involved into study and develop new detection algorithms and investigates deep learning based object tracking techniques for this specific case. The student is expected to be familiar with Python and Tensorflow.

Assistant: Saeed Rad ([email protected])
Supervisor: Prof. Jean-Philippe Thiran

11. Computer vision for the EPFL Racing Team

EPFL Racing team is a student team that competes in the Formula Student competition, the largest student competition for engineers in the world. We design and build electric race cars and compete against hundreds of other universities during summer, all across Europe.

Teams are scored on different ‘events’ such as car acceleration, endurance, agility and are driven by pilots that are students, members of the team. Performance wise, to give an example the 0 – 100 km/h is realized in approximately 2.8 sec!

Starting in summer 2021 we plan on having an autonomous car, able to visualize the track and to find the perfect line to get the best lap time! To do so, we are looking to start projects on the topics of computer vision.

The aim of the projects would be to find and implement the most performant and robust detection algorithms, knowing that the track is bounded by yellow and blue cones (cf. picture of the car of ETH in 2017). The car must be as quick as possible around the track while hitting as few cones as possible, otherwise we get a time penalty. Also, it will be crucial to consider the hardware implementation on the 2021 car: what sensors, where to put them and how much power is required.

Supervisor: Prof. Jean-Philippe Thiran