Simon Deniffel

I have been working on learning-based image coding since January 2024. The primary focus is to compress images using neural networks to reduce the data rate while maintaining high visual quality. More specifically, I investigate how the encoding process can be adaptively influenced based on image content to improve the compression rate. These approaches are independent of the chosen learning-based coder and represent an augmenting technique.

I offer theses in the field of learning based image coding. In particular, for the following topics:

• Hyperparameter search for adaptive quantization
• Network overfitting on single images
• Integration of new architecture approaches into JPEG AI
• Efficient training on very small datasets
• Optimization of the architecture of entropy models