Recent advances in computer vision have led to remarkable and even superhuman performance on tasks such as image prediction. 3-D scene understanding is more difficult with many parts of objects occluded or hidden. In this project, we seek to predict semantic labels and occupancy for voxels in rooms using shape primitives and recurrent convolutional networks from a single depth map of the scene.

I decided to participate in SuperUROP to get a more immersive experience in research. In the past, I have tried several deep learning-based projects in natural-language processing (NLP) computer vision and reinforcement learning. I am really excited to get a better understanding of where we are in 3-D computer vision and to hopefully improve our performance on 3-D computer vision tasks.