I build computer-vision and machine-learning systems for high-resolution imagery and multimodal data. My work runs from UAV data collection and geospatial processing to distributed PyTorch training on A100 GPUs and documented tools that other researchers can run.
I’m a Ph.D. student at the University of Florida with more than four years of experience in computer vision, machine learning, and geospatial processing for high-resolution imagery and multimodal sensor data.
I work across the full pipeline, from UAV data collection and sensor integration to image processing, model development, and field-based evaluation. My projects combine RGB, multispectral, thermal, hyperspectral, and elevation data, with individual flight collections often exceeding 30 GB.
My modeling work includes SAM, Depth Anything, YOLO, CNNs, LSTMs, TCNs, and Transformers. I train high-resolution models using CUDA and distributed PyTorch on NVIDIA A100 GPUs and maintain reproducible SLURM launch files, checkpoints, outputs, and experiment logs.
I also build tools for other researchers. These include a modular UAV phenotyping pipeline that transforms raw drone imagery into standardized plant- or plot-level traits, PlantSegFlow for zero-shot plant-video segmentation and tracking, and PhenomeInsight, a deployed web-GIS platform for processing and visualizing UAV data through UF HiPerGator.
+ DSSAT crop-modeling workshop — process-based simulation to complement the data-driven side.
Engineered a reproducible benchmarking framework that integrates computer vision, tabular learning, and environmental time-series modeling. The system standardizes data preparation, grouped cross-validation, fold-specific preprocessing, and out-of-fold evaluation across ML and deep-learning architectures, enabling fair comparison without data leakage.
The benchmark combines UAV imagery, elevation data, field measurements, management records, and daily fertilizer and weather sequences. Across two growing seasons, learned visual representations consistently outperformed handcrafted spectral and structural features. The modular design supports extension to other crops, sensors, imaging modalities, and spatial resolutions.
Designed, fabricated, and field-validated a modular UAV platform for synchronized RGB, multispectral, and thermal data collection. The system combined custom SolidWorks components and 3D-printed mounts with sensor integration, onboard power, wiring, data transfer, and storage. The published workflow converts the collected data into canopy height, coverage, and temperature traits validated against field measurements.
Co-authored a CVPR Workshops 2026 paper on a zero-shot vision pipeline that converts UAV and ground field videos into structured plant-level data. The system combines Video Depth Anything, SAM3, adaptive depth filtering, and motion-based identity assignment to segment plants, maintain row-aware ordering, track identities across frames, and generate representative images and trait-ready outputs for downstream breeding analysis.
Developed and maintain a modular pipeline that transforms raw drone imagery into analysis-ready phenotyping datasets. The workflow supports repeated collections and both plot- and plant-level studies, automating photogrammetry, orthomosaic and DEM generation, vegetation segmentation, spatial ID alignment, trait extraction, and standardized export for downstream breeding and research analyses.
Contributed to a deep-learning system for detecting strawberry runners in ground and UAV imagery. The model was trained and evaluated on a diverse, multi-platform dataset to improve robustness across imaging perspectives and field conditions.
Contributed processing and visualization features to PhenomeInsight, a deployed UF web-GIS platform that connects user-uploaded UAV data with HiPerGator. The application manages processing jobs and returns interactive orthomosaic, elevation, vegetation-index, hotspot analysis, and job-status notifications to users.
A growing shelf of little characters.
Behind the lens.