Learning Deep Representations from RGB-D data


Exploiting RGB-D data by means of convolutional neural networks (CNNs) is at the core of a number of robotics applications, including object detection, scene semantic segmentation, and grasping. We study novel approaches for learning estimating depth information from RGB data or from learning deep representations from RGB-D data.


Relevant Publications:
  • A. Pilzer, S. Lathuilière, N. Sebe, E. Ricci, Refine and Distill: Exploiting Cycle-Inconsistency and Knowledge Distillation for Unsupervised Monocular Depth Estimation, CVPR 2019.
  • A. Pilzer, D. Xu, M. Marian Puscas, E. Ricci, N. Sebe: Unsupervised Adversarial Depth Estimation Using Cycled Generative Networks. 3DV 2018.
  • L. Porzi, S. Rota Bulò, A. Penate-Sanchez, E.Ricci, F. Moreno-Noguer, Learning Depth-aware Representations for Robotic Perception, Robotics and Automation Letters (RA-L), 2017.