As an intern you will work on advanced multimodal computer vision problems, focused on improving visibility and robustness in low-light conditions by combining RGB and thermal imaging.
Tasks:
- Work with visible (RGB) and thermal image and video datasets, including internal and public sources
- Understand and reproduce baseline pipelines for image registration and fusion
- Design and implement an evaluation framework with quantitative metrics and visual assessment
- Implement and benchmark fusion algorithms for RGB-thermal fusion
- Design and train confidence-aware fusion models using attention or gated fusion
- Improve robustness for imperfect alignment and low visibility
- Develop methods to identify and minimize artifacts such as ghosting, hallucinations and distortions
- Evaluate results against baselines with performance metrics and qualitative insights
- Validation on downstream tasks such as object detection in low-light conditions (optional)
- Document findings in thesis/report and prepare publication-ready results package