Semantic Segmentation using Deep Learning and its Applications

Semantic scene segmentation is a challenging problem that has great importance in many applications, including assistive and autonomous navigation systems. Such vision systems must cope with image distortions, lighting variations, changing surfaces, and varying illumination conditions. In this talk, we will present deep learning-based vision systems for fast and accurate object segmentation and scene parsing. Furthermore, the talk will present a hybrid deep learning approach for semantic segmentation. The new architecture combines Bayesian learning with deep Gabor convolutional neural networks (GCNNs) to perform semantic segmentation of unstructured scenes. In this approach, the Gabor filter parameters are modeled as normal distributions with mean and variance that are learned using variational Bayesian inference. The resulting network has a compact architecture with smaller number of trainable parameters, which helps mitigate the overfitting problem.