Authors : Arindam Das , Saranya Kandan , Senthil Yogamani
Semantic segmentation remains a computationally intensive algorithm for embedded deployment even with the rapid growth of computation power. Thus efficient network design is a critical aspect especially for applications like automated driving which requires real-time performance. Recently, there has been a lot of research on designing efficient encoders that are mostly task agnostic. Unlike image classification and bounding box object detection tasks, decoders are computationally expensive for semantic segmentation task. In this work, we focus on efficient design of the segmentation decoder making use of an efficient encoder. We design a novel efficient nonbottleneck layer and a family of decoders which fit into a small run-time budget using VGG10 as efficient encoder. We demonstrate in our dataset that experimentation with various design choices led to an improvement of 10% from a baseline performance. The optimal configuration of the decoder leads to a performance of 50 fps on an embedded low power SOC namely Renesas V3H.
Keywords : Semantic Segmentation, Visual Perception, Embedded Systems, Automated Driving