Deep Learning for Image Synthesis
In this course you will implement deep learning methods in PyTorch. It focuses on problems from computer vision and image synthesis.
The course takes place in two phases:
- Self-study (around 2CP): Two weeks.
- Repetition of foundations (if you already know the basics, this will not be a lot of work)
- Implementation of different methods (for example, style transfer, image-to-image, diffusion models) in exercise notebooks.
- Multiple meetings to answer questions.
- Project phase (7CP): Variable length, depending on workload.
- Work on a research project within our group. The work can be a starting point for a bachelor thesis in our group.
- Weekly meetings with a supervisor.
- Finally, you present the result of you project in a 10min presentation in our lab meeting.
As a first step contact a potential supervisor from our group that offers a project that you find interesting (see the list of thesis offers below). Next, you agree on a starting date and time plan with together with your supervisor and the course coordinator. The course can be done individually but if there are several students interested we will setup group meetings during the self-study phase.
Thesis Offers
The following is a list of thesis offers. Often, these topics can also be used as a project topic for this course:
2D Clustering with Computer Vision
Using learned models for instance segmentation to cluster 2D data
Supervisor: Martin Ritzert
Animal Re-Identification on the PetFace dataset
Explore performance and transfer learning capabilities of re-ID methods on a new dataset.
Supervisor: Felix Benjamin Müller
Automatic Detection, Segmentation and Classification of Lemur Vocalizations
Supervisor: Dr. Kaja Wierucka (external)
Data-centric learning for primate action recognition
Building intelligent filtering of large-scale video data to improve model performance
Supervisor: Felix Benjamin Müller
Design Optimization for Acoustics
Using Neural Networks to predict the frequency responses on beading patterns.
Supervisor: Jan van Delden
Embeddings for neurons function and how they relate to morphology and cell types
Improve functional neuronal clustering
Supervisor: Polina Turishcheva
Generative Deep Learning for Lidar Scans of Trees
Flow matching for 3D structure generation with the application of lidar scans of trees
Supervisor: Jan van Delden
Implicit_learning_for_neuronal_prepresentation
Improve functional neuronal clustering
Supervisor: Polina Turishcheva
Mamba-inspired model for neuron-to-neuron architecture
Improve functional neuronal clustering
Supervisor: Polina Turishcheva
Model Neurons Interactions in time and between each other
Adjust readouts for neuroscience vision models to consider time and neurons interactions
Supervisor: Polina Turishcheva
Perspective module to module eye focus for mouse visual cortex
Adjust the module from the foundational model and try to make it sharable between animals
Supervisor: Polina Turishcheva
Solving Citation Networks with Large Language Models
By focussing on the abstract, LLMs should be able to effectively solve Cora and other citation datasets
Supervisor: Martin Ritzert
Tracking Nasal Skin Temperature in Free-Ranging Macaques
Supervisor: Alexander Ecker, Julia Ostner (external)
Treewidth-Based Positional Encodings
Generalizing a positional encoding for trees to general graphs using the notion of treewidth
Supervisor: Martin Ritzert
Using Video Backbone Models for Monkey Detection and Tracking Models
Can we make use of the representations learned by large-scale backbones for monkey detection?
Supervisor: Felix Benjamin Müller
