Deep Learning Researcher & Engineer — Computer Vision · NLP · LLM Agents
Deep learning across vision and language — from generative vision pipelines to LLM agents for real-world deployment.
M.Sc. Computer Science student at TUM (Technical University of Munich), researching computer vision, NLP, and deep learning for 3D geometry. Currently a Student Assistant at the TUM Computer Vision Group (MuMoL), building LLM agents and adapting open-source models for medical deployment.
🔭 Currently: joining Sophia Koepke and Alexei (Alyosha) Efros at BAIR, UC Berkeley for a summer research stay.
F. Förster, Q. Khan and D. Cremers, "Decentralized Reinforcement Learning for Multi-Agent Navigation in Unconstrained Environments," 2025 IEEE Intelligent Vehicles Symposium (IV), Cluj-Napoca, Romania, 2025, pp. 2067-2073, doi: 10.1109/IV64158.2025.11097389
| Project | Description | Tags |
|---|---|---|
| Decentralized RL for Multi-Agent Navigation | Published at IEEE IV 2025 — decentralized reinforcement learning for multi-agent vehicle navigation in unconstrained environments. | Reinforcement Learning Multi-Agent Systems |
| VBF Higgs-Pair Event Classifier | Benchmarking 12+ ML methods, including DeepSets, to classify vector-boson-fusion Higgs-pair events by coupling strength — 99.1% accuracy on sets of 10 events. | Machine Learning Particle Physics |
| STYLO-Pipeline | Vision-foundation-model pipeline for virtual try-on and semantic outfit editing, chaining six models incl. SAM2 and StableVITON. | Computer Vision Generative AI |
| pixelNeRF Upsampling Speedup | Cut pixelNeRF inference time ~19x (2.44s → 0.13s) by rendering downsampled images and recovering detail with a learned CNN upsampler. | Computer Vision 3D Geometry |
| Playing Card Object Detection | Flutter app that detects playing cards on-device with a fine-tuned, int8-quantized YOLO model and computes win probabilities via Monte Carlo simulation. | Computer Vision On-Device ML |
| Building Materials Segmentation | U-Net segmenting air voids, aggregate, and cement paste in concrete micrographs, trained with active learning and synthetic compositing. | Computer Vision Semantic Segmentation |
More details and write-ups at spatenfe.github.io.
Machine Learning & Deep Learning
Programming & Tools
