Professional Experience

Developing foundation-model-based imaging systems for Samsung smartphone cameras with emphasis on faithful, mobile-friendly super-resolution.

• • Developing foundation-model-based single-image super-resolution systems for Samsung smartphone cameras using diffusion architectures with VAE, U-Net, and vision transformer components, targeting high-fidelity restoration with minimal hallucinations on real camera content.
• • Organized large-scale image and text data for model development and worked closely on feature-driven training workflows, helping integrate DINOv2-based conditioning pipelines to replace text-only guidance and improve patch-level texture fidelity, robustness, and real-world generalization.
• • Researching flow-matching objectives and model condensation strategies for mobile deployment while contributing image-quality algorithm blocks to Samsung's Expert RAW pipeline.
• • Supporting published research on real-world super-resolution foundation models, including F2IDiff, with a practical focus on controllable generation for camera pipelines.

Applied computer vision and deep learning to degraded long-range video across ground, aerial, handheld, and satellite platforms.

• • Applied deep learning and computer vision methods for object detection, activity recognition, uncertainty estimation, segmentation, enhancement, and video or image search on degraded long-range video.
• • Worked with multi-source imagery from ground, handheld, aerial, and satellite cameras to improve feature preservation and degraded-scene understanding.
• • Built and optimized pipelines in modern deep learning frameworks such as PyTorch and TensorFlow for practical large-scale experimentation.
• • Developed an automated pipeline for DeepFake detection, evaluating hundreds of models to distinguish AI-generated from real images and identify the optimal encoder for the project.

Built applied machine learning pipelines on spectral measurements for healthcare and biological sensing problems.

• • Developed ML models for Covid-19 classification using spectral data from saliva, bacteria, and buffer samples.
• • Built spectral unmixing pipelines using PARAFAC and domain-specific preprocessing for biological signal separation.
• • Improved component identification workflows relevant to drug detection and pathogen analysis.
• • Migrated core analysis functionality into Python to improve repeatability and deployment readiness.

Designed a physics-aware turbulence estimation pipeline using multimodal sensor data and extensive real-world field experiments.

• • Built a deep learning model for atmospheric turbulence estimation across varying focus distances, illumination, platform motion, and camera shake.
• • Managed onsite field experiments involving telescopes, drones, cameras, weather stations, and scintillometers.
• • Analyzed and processed multidimensional sensor data to connect visual degradation with physical turbulence measurements.
• • Translated the resulting system into published research contributions on structure index estimation and long-range imaging.

Conducting research at the intersection of computational imaging, photography, perception, and robust restoration under real-world degradation.

• • Designed physics-based deep learning models for dynamic scene restoration in ultra-zoom and astrophotography cameras, spanning image formation, degradation modeling, and restoration under real-world atmospheric turbulence.
• • Conducted research in computational imaging, photography, computer vision, and visual or perceptual quality with emphasis on robust evaluation.
• • Gathered, curated, and validated research datasets while building restoration, segmentation, and simulation pipelines for long-range imaging.
• • Authored papers, proposals, and technical reports that connect imaging theory with practical vision systems.

Worked on medical image analysis, multimodal deep learning, and infrared meibography for automated ocular assessment.

• • Designed a multimodal deep learning architecture for meibomian gland analysis using GANs, ResNet-50, and encoder-decoder networks.
• • Enabled automated assessment of infrared meibography by detecting and segmenting the gland region and removing specular reflections.
• • Contributed to a public dataset of 1,600 annotated infrared images for gland analysis and qualitative grading.
• • Collaborated on microscopy-related imaging research in confocal, diffraction, and light-sheet settings.