Research Project Title:
Machine Learning with Computer Graphics for Self-Driving Car
abstract:The state-of-the-art vision algorithm for self-driving cars uses deep learning, which requires a lot of data to train the models and to test their robustness in different environments. However, obtaining extensive photographs of streets and the corresponding ground truth annotations is extremely expensive and sometimes impractical. Our approach is using computer graphics to generate large annotated datasets by rendering scenes in different controllable lighting and environments. This opens up opportunities to study various aspects of deep learning, such as how well neural network models generalize, and how to make the learning invariant to lighting conditions. Another challenge is making the virtual world realistic while also optimizing the performance of the algorithm.
As a computer science student and photography enthusiast, I’m interested in researches where the two fields overlap. This SuperUROP is a great opportunity for me to gain more experience in machine learning and computer graphics research. I hope to improve my understanding of these topics, to learn how to collaborate on a successful project, and to make an impact with real-world applications.