Pietro Bonazzi

Our paper on fully ternary Vision Transformers for microcontrollers has been accepted at NeurIPS 2026. FTerViT achieves 82.43% ImageNet-1K top-1 accuracy at just 6 MB, with the first deployment of a ternary ViT on an ESP32-S3.

Presented a poster at the 3rd EFCL International Workshop on Sustainable AI at ETH Zürich, bringing together researchers and industry experts on efficient, scalable, and responsible AI systems.

Presented our work on TinyGLASS, a real-time self-supervised in-sensor anomaly detection system, at the 8th IEEE International Conference on AI Circuits and Systems (AICAS) in Ha Long Bay, Vietnam.

Selected to attend Y Combinator's Startup School 2026 in San Francisco (July 25–26), a gathering of the world's most promising builders and founders.

Selected to attend the 20th edition of the International Computer Vision Summer School (ICVSS 2026) in Sicily, Italy (July 5–11), themed 'Computer Vision for Spatial and Physical Intelligence'.

PicoSAM2 and PicoSAM3, ultra-lightweight segmentation models running in-sensor on the Sony IMX500, accepted for publication in the IEEE Sensors Journal.

Our team won the ETH Kite Award 2026 for the embedded systems course.

Attended the World Economic Forum Annual Meeting 2026 in Davos, Switzerland (January 19–23), representing the ETH Zürich startup ecosystem at the USA House.

Received the Outstanding Lecture Award at IEEE Sensors 2025 in Vancouver for our presentation on in-sensing image segmentation.

Our team won the G-Research Quant Challenge at ETH Zürich, a quantitative finance competition focused on cointegration-based trading strategies tested on real-world datasets.

Stefan Zihlmann, one of my students, was awarded for outstanding work on PicoSAM2, a low-latency segmentation system running directly on the Sony IMX500 camera sensor.

At IEEE's Workshop on Advanced Sensors & Interfaces in Puglia, our research on event-based eye tracking using the STM32N6 was awarded Best Paper.

We'll be presenting our FPGA-accelerated obstacle avoidance system using event cameras at the EventVision Workshop during CVPR 2025.

Spent 10 days touring Jiangsu, Zhejiang, and Tianjin, unveiling our smart-glasses work and exploring innovation clusters across China's Yangtze River Region.

Thrilled that our smart glasses object detection work has been accepted for presentation at ICVSE during ECCV 2024 in Milan.

Proud to introduce Retina, our novel event-based eye tracking system and open dataset, at the 1st Workshop on AI for Streaming at CVPR 2024.

Our paper was selected for oral presentation at the 6th IEEE AICAS, held at Khalifa University in Abu Dhabi.

Thrilled to be joining the EF 2024 Cohort! I'll be learning valuable skills to validate entrepreneurial ideas and test co-founder partnerships.

Anthropic's CEO discussed TinyissimoYOLO on ImportAI (Jack Clark's newsletter). We'll keep working on the final system to open source hardware and software!

Great to see Nick Bild's insightful analysis of TinyTracker featured on Hackster.io!

Excited to announce we received a 2k Fr. Grant from InnovationNetwork.ch!

Honored to be invited to the Gala organized by Bocconi University for being included in the Nova Talent Italian List 2022 for Computer Science.

Excited to share my work on few-shot point cloud reconstruction and denoising using learned Gaussian splats and fine-tuned diffusion features, part of the Robotics & Perception Group (UZH) and Google Inc.

Proud to present our project on 3D scene generation from scene graphs and self-attention, developed with the Robotics & Perception Group (UZH) and Google Inc.