In my role as a Software Engineer Intern at A*STAR's I²R, I leveraged my skills from MIT, particularly in computer science and machine learning, to quickly adapt and contribute to the project. Although I initially had no experience with neural networks or relocalization, I utilized my ability to learn new skills rapidly and think creatively. My main project involved researching methods to enhance relocalization using machine learning techniques. Key contributions included:
- Conducting performance analyses of seven novel feature extraction heads for encoding prior to a Perspective-n-Point (PnP) head.
- Investigating ways to enhance the performance of these feature extractors by integrating them with semantic segmentation models.
- Refining the extractors through data augmentation and hyperparameter optimization.
