Computer Vision & AI Internship Candidate

NGUYEN MINH HUY

PROFILE

Third-year Computer Science student at Ho Chi Minh City Open University with a 3.5/4.0 GPA, focused on Computer Vision and Deep Learning. Built research-oriented and real-time vision systems using PyTorch, DINOv2, YOLOv11, TensorRT, and diffusion models. Seeking an AI/ML or Computer Vision Engineering Internship to apply and deepen practical machine learning and computer vision skills.

EDUCATION

Ho Chi Minh City Open University

Relevant Coursework: Computer Vision, Advanced Artificial Intelligence, Machine Learning, Data Mining, Mathematics for Machine Learning, Probability and Statistics, Data Structures and Algorithms, Object-Oriented Design Patterns. Ho Chi Minh City, Vietnam

Bachelor of Science in Computer Science; GPA: 3.5/4.0 Sep. 2023 – 2027 (Expected) PROJECTS

Fashion Try-On Pipeline Deep Learning, Stable Diffusion, IP-Adapter, PyTorch Jun. 2026 – Jul. 2026 Built an end-to-end 2D virtual try-on pipeline integrating SegFormer, IP-Adapter, Stable Diffusion Inpainting, and PEFT LoRA for automated masking and garment-conditioned generation. Developed a FastAPI inference API with a responsive web interface for person/garment image inputs. Evaluated on 2,032 Zalando VITON test pairs, achieving 0.7427 CLIP garment alignment score, 0.8523 SSIM, and 8.74 FID, with 91.49 ms latency (~10.9 FPS).

Repository: https://github.com/MinhHuy128/Fashion-Try-On (Personal Project) Real-Time Traffic Sign Detection PyTorch, YOLOv11, TensorRT, SAHI Mar. 2026 – Jun. 2026 Engineered a multi-backend real-time vision pipeline with TensorRT FP16, ONNX Runtime, and PyTorch to detect and track 56 Vietnamese traffic sign classes from dashcam video. Addressed small/distant signs and temporal prediction instability by integrating Adaptive SAHI, ByteTrack, and a custom 1D Kalman voting stabilizer into the detection-tracking pipeline. Evaluated both detection quality and inference efficiency, achieving 79.52% mAP@0.50:0.95, 95.99% mAP@0.50, 91.44% precision, and 90.86% recall, while reaching 6.04 ms mean latency (165.5 FPS) with TensorRT FP16, single-frame inference. Repository: https://github.com/MinhHuy128/Sign_Detection_VN (Personal Project) BayesIntent: ATIS Intent Classification Python, Naive Bayes, NLP, FastAPI Led a 3-member team to build a Multinomial Naive Bayes classifier from scratch with Laplace smoothing & log-space inference, achieving 89.26% test accuracy (89.33% 5-fold CV) on ATIS. Built a FastAPI backend with RAM-cached parameter lookup and a vanilla JS web UI for real-time log-score visualization. Organized team workflows via Agile sync-ups and decoupled architecture for parallel development, while managing Git Flow branching and conducting code reviews.

Repository: https://github.com/MinhHuy128/BayesIntent-ATIS (Team Project - 3 Members) Apr. 2026 – May 2026

SKILLS

Languages: Python, C++, SQL, C#, Java

AI & Machine Learning: PyTorch, OpenCV, DINOv2, Vision Transformers, YOLOv11, Stable Diffusion, IP-Adapter, SegFormer Tools & DevOps: Git, Linux, Docker

Others: Agile/Scrum, Project Management, Team Leadership, Problem Solving LANGUAGES

Vietnamese: Native English: B2

CERTIFICATIONS & VOLUNTEER EXPERIENCE

Certificate: Advanced AI Training Program (Samsung Innovation Campus) Volunteer: Core Organizer for Cam Chi Orphanage Charity & Spring Volunteer Campaigns Apr. 2025 – Sep. 2025

2024 – 2026

+84-939-***-*** ️ ****************@*****.*** https://www.linkedin.com/in/minhhuy128

https://github.com/MinhHuy128

Industrial Anomaly Detection PyTorch, DINOv2 ViT-B/14, Vision Transformer Jul. 2026 – Present Developed an unsupervised anomaly detection framework for MVTec LOCO AD, combining a frozen DINOv2-Register ViT-B/14 encoder, Bottleneck MLP, and 8-layer Transformer Decoder with O(N) Linear Attention. Designed a Global Consistency Token (GCT) supervised by frozen DINOv2 CLS embeddings, with active dual-stream scoring to combine local patch errors and global consistency. Improved mean AUROC to 86.68% (+2.01 pp over Dinomaly), including 80.33% logical and 93.04% structural AUROC, while maintaining 11.6 FPS and 70.10% sPRO on MVTec LOCO AD. Repository: https://github.com/MinhHuy128/industrial-anomaly-detection (Academic Project)

Contact this candidate