I'm a Computer Science graduate, currently a Master's student in Computer Science at the Department of Computer Science, Federal University of Minas Gerais (DCC/UFMG), affiliated with Laboratory for Computer Vision and Robotics (VeRLab).
My research sits at the intersection of:
Computer Science → Computer Vision → Video Coding / Image Processing
→ Feature Matching → Efficient AI → Edge / Embedded AI → Research Engineering
I work on making learned vision systems, from local feature matchers, understand, run, and hold up under real computational constraints: limited memory, quantized weights, and NPUs instead of GPUs.
Feature Matching on Embedded Hardware (XFeat on the edge) Investigating the deployment of XFeat, a lightweight local feature extraction/matching network, on extremely resource-constrained hardware, the Luckfox Pico (Rockchip RV1103), running inference on its NPU without relying on a GPU. This involves the full pipeline and its failure modes: PyTorch → ONNX → RKNN conversion, INT8 quantization, operator compatibility, memory limitations, and numerical behavior across stages. I run systematic experiments comparing PyTorch, ONNX, and on-device NPU outputs to understand why specific components of the model degrade after quantization and deployment.
Feature Matching Research (broader line of work) This project is evolving into a wider research direction on efficient feature matching, covering local feature detection/description, contrastive and information-theoretic perspectives on feature selection, robustness under quantization, and hardware-aware neural network optimization.
Video Coding / Versatile Video Coding (VVC) Prior research line Earlier research on VVC, built on VTM 23.10 + 360Lib 13.6, focused on ML-assisted intra prediction for 360° video and transform-coding optimization (DST-VII / DCT-VIII, Multiple Constant Multiplication). This work produced FastMIP-360, achieving roughly 5.74% encoding time reduction at approximately +0.171% BD-BR, and led to accepted publications at LASCAS 2025 and WebMedia 2024. This line laid the groundwork for my current focus on efficient, hardware-aware vision systems.
Languages
Python C++ C Bash
AI & Machine Learning
PyTorch ONNX TensorFlow scikit-learn NumPy
Computer Vision
OpenCV scikit-image XFeat YOLO DeepSORT · feature detection/description/matching · object detection · tracking · image processing
Efficient AI / Edge AI
RKNN · Rockchip NPU (RV1103) · INT8 quantization · ONNX deployment · embedded inference · hardware-aware optimization
Video Coding (VVC)
VTM 360Lib · intra prediction · transform coding · Multiple Constant Multiplication (MCM)
Backend & AI Infrastructure
FastAPI Flask SQLite ChromaDB Ollama · local LLMs · RAG
Research Engineering & Tools
Docker Linux Git GitHub VS Code GitHub Copilot · experiment automation · benchmarking · reproducible research
📫 Feel free to reach out:
- 🌐 Portfolio
- 📜 Currículo Lattes
- 🆔 ORCID
