Daniel Rossi

I am a final-year PhD Candidate in Artificial Intelligence at the University of Modena and Reggio Emilia, working at AImageLab under the supervision of Prof. Roberto Vezzani.

My research focuses on Efficient Deep Learning and Edge AI deployment. I design ultra-lightweight neural networks for computer vision, and I study how they actually behave once they leave the workstation: latency, energy per inference and accuracy on NPUs, embedded GPUs and microcontrollers. Application domains include aerial disaster response, surveillance, on-device fine-grained recognition and monocular depth estimation.

Recent work includes TakuNet (WACV 2025 Workshops), an energy-efficient CNN reaching more than 650 fps on a 15 W Jetson Orin Nano, its journal extension in Image and Vision Computing, and BoltNet (CVPPA @ ECCV 2026), a 341K-parameter network for on-device plant species identification benchmarked across CPU, GPU and NPU. I also maintain open-source tooling for this line of work, such as AutoDock and EdgePowerMeter.

Previously, I received my Master’s Degree cum laude with a thesis entitled “3D Human Pose Estimation from RGB Images with Transformers and Deformable Attention”. During this period, I also had the opportunity to work for 6 months as an intern at Toyota Motor Europe in Brussels, focusing on NeRF, Gaussian Splatting, 3D Pose and Hand Estimation.

Outside the lab I run Projecto, a YouTube channel about electronics, embedded systems and deep learning.