Research Interests

My main research is computer vision: image retrieval, sparse and dense feature matching, robust geometric estimation (RANSAC/MAGSAC++), segmentation, and diffusion-based synthesis. I apply these methods to UAV visual geo-localization without GPS, multimodal video retrieval, medical imaging and generative AI — see the Publications page for the papers themselves.

CNN-based ViT-based CLIP-based R-CNN-based RANSAC-based Diffusion-based Feature Matching
UAV Visual Geo-Localization

Geo-locating a UAV from a single nadir image against a satellite map without GPS — via geometry-based building-footprint matching (Delaunay triangulation + KD-tree) for an undergraduate research project, and via coarse-to-fine cross-view retrieval and feature matching (XFeat, RoMa) with RANSAC/MAGSAC++ verification for my undergraduate thesis.

Multimodal Video Retrieval

Text-to-keyframe search over large video collections, combining vision-language embeddings (BEiT-3, CLIP, Perception Encoder) with vector databases (FAISS, Qdrant) and cross-modal, temporal multi-event query processing.

Medical Image Segmentation & Synthesis

Training-free statistical-graph segmentation (KDE-seeded GMM-EM + 3D Graph Cut) for brain-tumor MRI, and label-conditioned latent diffusion (Stable Diffusion + LoRA) for pathology-specific chest X-ray synthesis.

Generative AI

Diffusion and flow-matching models, including co-speech gesture generation synchronized with audio, built on the mathematical foundations that make feature extraction and sampling reliable.

Currently

See the full CV for dates and detailed responsibilities at each lab.