ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation

The ability to perform pixel-wise semantic segmentation in real-time is of importance in mobile applications. Recent deep neural networks aimed this task have the disadvantage of requiring a large number of floating operations and have long run-times that hinder their usability. In this, we propose a novel deep neural network architecture named ENet (efficient neural network), created specifically for tasks requiring low operation. ENet is up to 18\ faster, requires 75\ less, has 79\ less parameters, and provides similar or better accuracy existing models. We have tested it on CamVid, Cityscapes and SUN datasets report on comparisons with existing state-of-the-art methods, and the-offs between accuracy and processing time of a network. We present measurements of the proposed architecture on embedded systems and possible software improvements that could make ENet even faster.