Five interconnected research areas that form the foundation of my work — from fundamental computer vision to applied industrial systems.
First-person action recognition & understanding from wearable cameras. Bridging human activity analysis with computer vision in unconstrained environments.
Temporal video understanding and classification. Designing architectures that capture motion dynamics across diverse domains and perspectives.
Building AI systems that generalize across domains without retraining. Unsupervised and semi-supervised transfer learning for cross-domain robustness.
Identifying rare, unusual or dangerous events in industrial and video streams. Unsupervised and one-class approaches for zero-shot anomaly localization.
Pixel-wise classification for dense scene understanding. Applications in autonomous systems, industrial inspection and AR interfaces.
Integrating audio, visual and sensor signals for comprehensive scene understanding. Joint representation learning across modalities.
Real-time and high-accuracy object detection pipelines for unconstrained scenes. From anchor-based to transformer-based detectors deployed in production.
Fine-grained visual recognition of objects across categories, instances and contexts. Metric learning and embedding-based approaches for few-shot and open-set scenarios.
Skeleton-based and heatmap-driven pose estimation for activity monitoring, action anticipation and human-machine interaction.
Visual quality control with anomaly detection — near-zero false positives on production lines.
Turning raw sensor and video streams into structured insights — dashboards, KPIs and predictive analytics.
Lightweight and quantized models optimized for mobile and edge hardware — Android & embedded systems.
Not many.
But every single one cost
blood, sweat & deadline nights.
— quality, not quantity.
Two-time consecutive winner of the Unsupervised Domain Adaptation track — Top 3 for two years running.
From hand-crafted descriptors to foundation models — one backbone to rule them all.
Presented cutting-edge first-person action recognition to Italy's Python AI community.
Open to research collaborations, industry projects and speaking engagements. Currently at ARGO Vision · Politecnico di Torino · IIT.
