A Semi-automated Annotation Tool using Segmentation and Multi-modal Data
Segmentation which divides a scene into subparts while preserving their local structures provides the way to understand a scene image more naturally and efficiently. One prominent example of segmentation is superpixel segmentation which has wide applications in image and video processing. For general scene representation different scene components can be captured from multi-modal data. These multi-model data give us various information including geometry and appearance of a scene. Our super urop project will develop a semi-automated annotation tool that labels scene objects when scene images sensor data and partial annotations are given. The annotation tool will annotate unlabeled superpixels in a scene from a generative model on segmentation inferred from multi-modal data.

I am interested in building mathematical model that explains complex structure of data. I believe our project which requires understanding of mathematical model and efficient data processing is the best fit for me to learn a lot. I hope to learn a lot about object annotation through various probabilistic models.