Cauã M. Pereira
I research computer vision & deep learning.
Computer Science undergraduate at UFMG and previous visiting researcher at the University of Alberta, working on visual representation, image matching, and multimodal perception.
Hi, I'm
About Me
I’m a Computer Science undergraduate at the Federal University of Minas Gerais (UFMG) with a strong background in research and hands-on experience in computer vision and deep learning. My journey in research began in high school, when I interned at LITE (Laboratory of Informatics, Telecommunications, and Electronics at UFMG). Since then, I’ve continued working on research projects at VeRLab (Computer Vision and Robotics Laboratory).
I’m currently working on a project with Petrobras, the largest oil and gas company in Latin America, focused on semantic segmentation of 3D point clouds. In parallel, I’m researching reinforcement learning and reference conditioning for keypoint detection, as well as non-rigid synthetic data generation for training data.
I’m always looking to collaborate on projects that push the boundaries of what’s possible, while learning, contributing, and building meaningful solutions.
Research Interests- Computer Vision
- Deep Learning
- Image Matching
- Multimodal Perception
- Visual Representation
Experience
Student Researcher — Image Matching
-
VeRLab
Sep 2025 - Present
- Contributed to the ideation and experimental design of a novel keypoint descriptor for image matching, currently under review at NeurIPS 2026.
- Led benchmarking across five established benchmarks; the proposed method outperformed strong baselines on the majority while staying competitive on the rest.
- Extending the direction with reinforcement learning and reference-conditioning for keypoint detection, plus non-rigid synthetic data generation for training.
Student Researcher — Petrobras Multivisão 3 (3D Semantic Segmentation)
Aug 2025 - Present
- Adapted state-of-the-art point cloud models for complex real-world industrial environments.
- Built scalable pipelines for large-scale point cloud partitioning, dataset preparation, and performance optimization.
- Optimized model architectures to improve segmentation accuracy and robustness under challenging 3D conditions.
- Achieved state-of-the-art recognition (under 5% WER) on the recognition component of a multimodal, bidirectional sign language–speech translation system bridging communication between Deaf patients and healthcare professionals.
- Implemented a complete data pipeline (preprocessing + training acceleration), reducing training time by over 80% and improving generalization across users.
- Expanded and maintained a custom Libras dataset by coordinating volunteers and conducting new video recordings.
- Designed and optimized deep learning architectures (ViTs, CNNs, RNNs) with targeted data augmentation, and mentored a new lab member.
Visiting Researcher — Computer Vision
-
Vision and Learning Lab — University of Alberta
Jan 2026 - Apr 2026
Selected for the University of Alberta Research Experience 2026, a competitive international research program.
- Researched multimodal semantic segmentation, adapting deep learning models and data pipelines to fuse RGB and thermal (multispectral) imagery.
- Reviewed the state of the art in domain adaptation and its application to multispectral semantic segmentation.
- Built reproducible large-scale training pipelines on the Alliance of Canada’s HPC clusters using Slurm for GPU job scheduling and orchestration.
Student Researcher — IoT and Embedded Systems
-
IHM Stefanini — LITE
May 2023 - Apr 2024
- Developed a module that collects data from electrical and magnetic sensors and transmits it in an unstructured format over LoRa, using a custom multi-point-to-point protocol based on CSMA/CA, delivering data to an Azure container for further analysis.
- Worked at LITE (Laboratory of Informatics, Telecommunications, and Electronics of UFMG) as a Technological Initiation and Innovation scholarship holder on the Smart Industrial Pumps Monitoring (MoBI IoT) project, collaborating with the Engineering Director at IHM.
- Contributed to SADUMBA and FIFO2 — both master’s projects — enhancing IoT data transmission for dam monitoring and biomedical applications, respectively.
Electronics Intern — IoT
-
LITE
Jun 2022 - Dec 2022
- Built from scratch a system for the Smart Maturation Chamber of Minas Artisanal Cheese project (CMI-QMA IoT) — part of a Doctoral Technological Innovation initiative — implementing real-time environmental data acquisition and encrypted transmission.
- Assisted students with IoT projects for the microcontrollers discipline.
Education
2020 - 2022
Electronics Technician (Integrated High School)
COLTEC — UFMG
Integrated High School and Electronics Technician program at the Technical
College of the Federal University of Minas Gerais.
Publications
INTERACT 2025
Captar-Libras: A Bidirectional Translator with a Photorealistic Avatar
Santos, N. S. et al. (incl. C. M. Pereira). Medical pre-consultation translator for Deaf patients. INTERACT 2025.
Software Registrations
MoBI IoT Project — 7 modules (INPI/UFMG, 2024)
Industrial pump monitoring via LoRa, MQTT, and cloud infrastructure. Da Cunha, A.B., Reis, M.S., Pereira, C.M. et al.
SADUMBA Project — 7 modules (INPI/UFMG, 2023)
Dam monitoring via ultrasonic and vibrating wire sensors. Da Cunha, A.B., Bastos, W.S., Pereira, C.M. et al.
FIFO2 Project — 2 modules (INPI/UFMG, 2023)
Biomedical flowmeter control and visualization. Da Cunha, A.B., Santos, M.J.C., Pereira, C.M. et al.
Get in Touch
My inbox is always open.