Building the bridge between State-of-the-Art Research and Production Systems.
Focusing on 3D Vision, Medical AI, and Efficient Deep Learning.
I am a Computer Vision & Machine Learning Research Engineer with a strong background in Software Engineering.
I don’t just train models — I design evaluation frameworks, benchmarks, and pipelines that make research results reproducible, interpretable, and deployable.
- 🔭 Current: Geometric Deep Learning & Diffusion Autoencoders for 3D face reconstruction @ MICC (University of Florence)
- 🧬 Medical AI: Designed and deployed chromosome recognition systems used in hospitals
- ⚡ Green AI: Studying energy efficiency of Deep Learning across languages, frameworks, and hardware
3D Vision · Geometric Deep Learning · Cross-Topology Modeling
- Developing DiffusionNet-based autoencoders for 3D face reconstruction
- Tackling cross-topology alignment (BFM ↔ FLAME ↔ FaceVerse)
- Designing identity-aware geometric metrics beyond Chamfer/L2
Evaluation Pipelines · Large-Scale Benchmarking
- Architected FaceBench, a modular framework for 3D reconstruction evaluation
- Executed 10k+ mesh comparisons with automated parallelization (~10× speedup)
- Proposed identity-aware metrics outperforming pure geometric distances
Green AI · System Efficiency
- Co-authored Green AI studies on DL energy efficiency
- Ran 60+ controlled experiments across 6 languages & 3 frameworks
- Observed up to 4.6× energy and 11× runtime variance
Medical AI · Production Systems
- Built chromosome detection systems with 95% accuracy on clinical data
- Deployed end-to-end pipelines (PyTorch → ONNX → C++)
- Reduced clinical analysis time from ~5 minutes to ~15 seconds
| Status | Title | Venue |
|---|---|---|
| ✅ | Green AI: Which Programming Language Consumes the Most? Marini, Pampaloni, Di Martino, Verdecchia, Vicario |
ICSE / GREENS 2025 |
| 🟡 | Deep Green AI: Energy Efficiency of Deep Learning across Programming Languages and Frameworks Pampaloni, Pagliocca, Vicario, Verdecchia |
Under Review |
| 🟡 | Evaluating the Energy Consumption of Kubernetes Clusters Verdecchia, Lago, Pampaloni |
Under Review |
| 🧊 | Cross-Topology Diffusion Autoencoders for Robust 3D Facial Reconstruction Pampaloni et al. |
In Preparation |
Medical AI — Production-Grade
- Deep learning framework for chromosome classification in the wild
- Handles overlaps, folding, and noisy acquisitions
- 95% accuracy on clinical data
- Tech: PyTorch, OpenCV, NumPy
Large-Scale 3D Evaluation Framework
- Modular pipeline for evaluating 3D face reconstruction methods
- Identity-aware metrics, topology handling, automated parallel execution
- Backbone of WBES and identity consistency studies
- Tech: Python, PyTorch3D, Trimesh
System-Level AI Efficiency Research
- Benchmarking training & inference energy cost
- Cross-language comparison (Python · C++ · Rust)
- Power tracking via NVIDIA-SMI and RAPL
| Area | Technologies |
|---|---|
| Languages | Python, Rust, Java |
| Deep Learning | PyTorch, PyTorch3D, TensorFlow, JAX, ONNX |
| 3D & Geometry | Open3D, Trimesh, LibIGL, ICP |
| Data & Analysis | NumPy, SciPy, Pandas |
| Tooling | Git, LaTeX, Linux |
