Eindhoven, Netherlands
Internship Explorer
Internship Explorer
In this project, you will explore how to combine modern deep learning with event-based and embedded processors to push the limits of what AR glasses can do on-device. You will help answer questions such as: * How can we architect models that are both accurate and ultra-efficient for real-world AR tasks on event-driven or low-power hardware? * What are the right trade-offs between accuracy, latency, memory, and energy for different AR scenarios? * How do we turn promising research ideas into practical, measurable improvements on realistic platforms and workloads? Your work will directly inform how future AR experiences can run locally, responsively, and efficiently on next-generation devices As a thesis student, you will define and drive a focused research direction in efficient on-device ML for AR, with a particular emphasis on event-driven or embedded processors. Possible directions within this space include: * Design and prototype ML models tailored to AR use cases under embedded constraints (e.g., event-based vision models, lightweight CNNs/Vision Transformers, or hybrid frame+event pipelines). * Set up datasets and baselines relevant to AR tasks (e.g., detection, tracking, segmentation, gesture/interaction), and define evaluation metrics across accuracy, latency, memory usage, and energy. * Implement and train models in PyTorch, including data pipelines, training loops, and evaluation scripts that are easy to extend and reproduce. * Explore efficiency techniques such as sparsity, pruning, quantization (PTQ/QAT), or event-based representations, and study their impact on performance–efficiency trade-offs. * Profile models under embedded-like conditions using simulators, profiling tools, or edge accelerators to understand system-level behavior (e.g., FLOPs, latency, memory footprint, bandwidth). * Communicate your findings through ablation studies, a clear thesis report (and optionally a paper-style write-up), and a reproducible codebase with pre-trained checkpoints.
