Master Thesis or Doctoral School Projects
- Deep Learning for Nuclear Fusion
- Promoting Connectivity of Linear Structures in 3D Microscopy Images
- Uncertainty Estimation in Volumetric Image Segmentation
- Modeling People and their Clothes in Crowded Scenes
- Multi-task Active Learning
- 3D Pose Based Motion Correction for Physical Exercises
- Physically Constrained Deep Networks
- Iterative Human Pose Refinement via Energy Ascent
For further project offers, please contact members of the CVLAB directly.
Semester Projects (Bachelor and Master)
Most of the offered semester projects can be rephrased for a thesis and vice versa. Please contact us directly.
Administrative
Semester Projects (Bachelor and Master)
SIN and SSC students do one semester project during their Bachelor studies and one semester project during their Master studies.
Semester projects can be done in groups of two students.
Semester projects are worth 8 credits for Bachelor and 12 credits for Master.
Students must have the approval of the Professor in charge of the laboratory before registering for the given project.
Oral defense: within two weeks of the hand-in date.
Master Thesis Projects
Master Thesis Projects are started once the complete master program is finished and all the credits have been obtained.
Projects for SSC and SIN students should last 4 months at the EPFL or 6 months in the industry or in another University.
Master Thesis Projects must be done individually.
Master Thesis Projects are worth 30 credits.
Students must have the approval of the Professor in charge of the laboratory before registering for the given project.
Additional information
An active research field is the prediction of physics mechanisms degrading plasma confinement and performance, eventually leading to disruptions on Tokamaks. Disruptions are the ultimate consequence of highly coupled non-linear plasma physics processes, resulting in an abrupt loss of plasma current and confinement inducing huge electromagnetic forces and thermal loads on Plasma Facing Components and (…)
As in many areas of computer vision, deep networks now deliver state-of-the-art results for delineation tasks, such as finding axons and dendrites in 3D light microscopy images. Most of the existing approaches rely on convolutional networks to extract from images binary masks denoting which voxels belong to neurites and which do not. Unfortunately, they do (…)
Fig 1: Common mistakes made by U-Net (inside orange squares)Current state-of-the art volumetric image segmentation approaches achieve impressive performances. But still, they are not 100% correct and make mistakes as you can see in Figure 1. Therefore we would like to know when a network makes a mistake, so that a human or another model (…)
While modeling people wearing tight-fitting clothes is fast becoming a mature field, handling subjects wearing looser garments remains an open problem when it is to be done in everyday settings where people may hide each other, precise outlines are hard to estimate, and shadows often complicate matters. Our goal is therefore to develop robust and (…)
Many practical continuous minimization problems, such as aerodynamic optimization, are not amenable to gradient-based optimization methods because derivatives cannot be computed directly. We recently showed that it is possible to train a Neural Network regressor as a proxy to the numerical simulator and optimize the proxy function via Gradient-Descent. [ICML 2018 Paper]. For example, we (…)
DescriptionDifferent visual tasks are often strongly and obviously correlated. For instance, having surface normals simplifies estimating the depth of an image, knowing segmentation could help detect objects, etc. Our intuition implies the existence of a special structure among visual tasks. Extracting this structure would allow us to seamlessly reuse supervision among related tasks or solve (…)
With the rise of self-management for treatment of musculoskeletal disorders, people tend to exercise alone and without supervision. However it is dangerous to attempt these exercises without feedback, as it can be difficult to realize when one is performing the exercise incorrectly. This could lead to further injury. Therefore our goal is to design a (…)
DescriptionModern CNN-based human pose estimators are constructed as regression architectures with human pose prediction as output. This formulation does not naturally involve the confidence of the estimation. Another possible solution is to regress the pose refinement directly via the CNN [2], which gives slight improvement over the ordinary regression-based techniques. However, it still suffers from (…)