Curriculum Vitae

đź§­ Technical Profile

  • Core Computer Vision: image retrieval with CNN/ViT backbones, sparse and dense feature matching, robust geometric estimation (RANSAC/MAGSAC++), segmentation, detection/classification, and diffusion-based image synthesis.
  • Applied to: UAV visual geo-localization without GPS (coarse-to-fine retrieval and matching; geometry-based building matching), multimodal video retrieval, and medical imaging.
  • Track Record: Presented UAV geo-localization work at CTMIA 2026; co-first author of SOICT 2026 (in press) and SOICT 2025 papers; First Prize, Olympic AI HCMC 2025; Final Round, AI Challenge HCMC 2025 & 2026.

🎓 Education

Bachelor of Science in Computer Science
2023 – Expected 2027
  • GPA: 3.0/4.0
  • Relevant Coursework: Image Processing & Computer Vision, Deep Learning, Machine Learning, AI, Linear Algebra, Probability & Statistics.
  • Languages: English (intermediate; capable of reading and writing technical materials).

📚 Publications

Co-first author on two SOICT papers. Full abstracts and contributions on the Publications page.

MRI Segmentation via Training-Free Statistical-Graph Framework In press
SOICT 2026
Thanh-Tin Bui, Hoang-Hiep Mai, Van-Duc Tran, Anh-Khoa Nguyen-Van, Duy-Tan Nguyen, & Khanh-Loi Nguyen.
  • Unsupervised, GPU-free brain-tumor MRI segmentation (StaGra: KDE-GMM + 3D Graph Cut); Dice 78.1% / 49.3% / 53.5% on BraTS 2021 with no labels.
Enhanced Multimodal Video Retrieval System Published
SOICT 2025
Van-Thinh Vo, Minh-Khoi Nguyen, Minh-Huy Tran, Anh-Quan Nguyen-Tran, Duy-Tan Nguyen, Khanh-Loi Nguyen, & Anh-Minh Phan.
  • My contribution: a KDE–GMM algorithm that automatically selects the transition-score threshold for shot- and slide-transition detection in the video segmentation stage.

đź’Ľ Research & Professional Experience

Undergraduate Thesis – Coarse-to-Fine UAV Visual Geo-Localization
In progress
  • Image retrieval (coarse): cross-view UAV-to-satellite retrieval with CNN- and ViT-based backbones on DenseUAV.
  • Feature matching (fine): sparse (XFeat) and dense (RoMa) local feature matching of UAV–satellite image pairs.
  • Geometric verification: robust model fitting with RANSAC / MAGSAC++ to reject outlier matches and estimate the UAV position.
  • Benchmarking: UAV-VisLoc, AnyVisLoc and AirZoo (in progress).
Student Researcher – UAV Visual Geo-Localization
Dec 2025 – Present
  • Goal: co-developing a texture-independent method that geo-locates a UAV from a single nadir image against a geo-referenced satellite map.
  • Instance segmentation: building footprints from a pre-trained Mask R-CNN, decomposed by Constrained Delaunay Triangulation.
  • Shape descriptors & indexing: 5-D geometric triangle descriptors (interior + Maximal Free Cone angles) indexed in a KD-tree for O(M log N) matching.
  • Dissemination: presented "Structure-Based UAV Visual Geo-Localization via Delaunay Triangulation and Maximal Free Cone Angle Descriptors" at CTMIA 2026 (Aug 2026); UAV-VisLoc evaluation in progress. Programme
AI Research Intern – Robot Navigation
Apr 2026 – Aug 2026
  • Integrated Social Navigation, Urban Navigation and Social Walker modules into the Arena-RosNav simulator on ROS2; built simulation scenarios and data-collection workflows to evaluate robot navigation behavior.
Core Member & Teaching Assistant
Sep 2024 – Present
  • TA for Object Detection and SVM practical sessions (Jun–Sep 2025, Jun–Sep 2026); mentored a 5-student team on C-DM chest X-ray synthesis (see Projects).
Student Researcher – Co-Speech Gesture Generation
Mar 2026 – Present
  • Developing diffusion and flow-matching models that generate gestures synchronized with speech.

🏆 Other Computer Vision Projects & Competitions

Multimodal Video Event Retrieval
Final Round, AI Challenge HCMC 2025 & 2026
GitHub Link | Certificates (PDF): 2025, 2026 | Paper: SOICT 2025 (see Publications)
  • Text-to-keyframe search engine with BEiT-3/CLIP embeddings on FAISS (2025), upgraded to BEiT-3, PE-Core and Jina embeddings on Qdrant (2026), supporting cross-modal and temporal multi-event queries.
Mushroom Image Classification
First Prize, Olympic AI HCMC 2025
Organized by HUTECH & HCA, May 2025 | GitHub Link | Award announcement
  • Ensemble of EfficientNet-B6, EfficientNetV2-M and ResNet-101 co-trained via Deep Mutual Learning on 32Ă—32 images with multi-augmentation and mixed precision; 96.17% accuracy / 96.10% F1 on the private test.
C-DM: Label-Conditioned Chest X-ray Synthesis with Latent Diffusion
Project Mentor, 2026
Mentored a 5-student team | Project Page
  • Guided a Stable Diffusion v1.5 + LoRA pipeline with a learned multi-hot label encoder on NIH Chest X-ray; a classifier trained only on synthetic images reached 95% of the real-data macro-AUC (0.69 vs. 0.72).

🔬 Technical Skills, Certifications & Awards

Programming

Python C C++ Java LaTeX

CV Methods

CNN-based ViT-based CLIP-based R-CNN-based RANSAC-based Diffusion-based Feature Matching

Frameworks & Tools

PyTorch Git GitHub FAISS Qdrant Arena-RosNav