My current research direction is centered on reliable adaptive perception: building vision systems that remain useful when the test distribution shifts, classes are imbalanced, environments change, or deployment constraints become strict. I work across both foundational adaptation methods and real-world autonomy problems.

PhD Research - Reliable Adaptive Vision

I focus on methodology for robust adaptation in open and dynamic environments, with emphasis on:

• Test-time adaptation
• Continual learning
• Long-tailed recognition
• CLIP/VLM adaptation
• Calibration and priors
• Training-free and lightweight adaptation

This line of work aims to improve reliability under label scarcity, distribution shift, and low-overhead deployment settings.

Industry Research - Robust AD/ADAS Perception

In parallel, I work on practical perception systems for real driving scenarios:

• Multi-camera road and free-space perception
• Robust segmentation under lighting, weather, material, and domain shift
• Temporal consistency for stable frame-to-frame behavior
• Camera-aware generalization across sensor setups
• Deployment on embedded platforms such as Jetson Orin
• Safety-aware perception and failure detection

This translational perspective helps connect algorithmic advances with safety-critical automotive deployment.