The document describes a study on semantic segmentation using resource efficient deep learning. It details the creation of a semantic segmentation dataset for objects in the RoboCup@Work environment. Various versions of the dataset are created by combining similar object classes. The DeepLabv3+ model is used with the MobileNetv2 and Xception encoders for semantic segmentation. The Xception encoder achieves higher accuracy but MobileNetv2 provides a more resource efficient solution. Quantizing the models results in significant reduction in memory usage with a small drop in accuracy.