
Do I need to finetune my VLM?
Vision-Language Models (VLMs) represent a fascinating intersection of computer vision and natural language processing. The combination of a Vision Encoder with a LLM has sparked interest in the computer vision field to use its capabilities for zero-shot tasks, where using traditional methods lack. Although expensive for large data throughputs, the image + prompt input makes VLMs versatile tools for visual question answering, captioning and various other tasks. SmolVLM architecture (Image from https://huggingface....

Thesis Work: Multi-View Hand Motion Reconstruction
The goal of this thesis was to reconstruct realistic hand motions in the presence of noise and occlusions. In the first part of this work, we process multi-view RGB input sequences with off-the-shelf methods to get a robust, but noisy MANO motion estimate. For the second part we split the estimation of the global and local features. As datasets covering both diverse global and local hand motion do not currently exist, we learn the dynamics of the two parts in isolation....
Starting my own Blog
Welcome to my blog. Since I spent a lot of time with the latest and greatest lately, I want you to learn from this experience: Vision-Language Modelling research (strengths, bias, training) VLMs vs Computer Vision Saving not the $$$ but our planet …

s'Tagebuech - A Digital Diary from Hand-Drawn Sketches
The Idea Sketching during holidays is great. It makes me relax, process thoughts and pay more attention to my current surroundings. But… Presenting sketches digitally has always been a struggle. My analog sketches require lots of lightning post-processing, cropping and often lack consistency. Sketches can be rough, approximate and incomplete. Mostly it’s enough to capture the impression. For my ultrarunning journey in Italy, I wanted to build something new. A lightweight application to tell a story with my drawings....

DOG at Home
Motivation I was always the kind of person who started a project only to find out that there is a much better idea around the corner that will blow everyone’s mind. The cycle kept repeating. With DOG at home I challenged myself to build a full clone of the popular swiss board game DOG during Covid in March 2020. It was different compared to previous projects, because there was direct use and need....