6. Conclusion
In this chapter, the concept of segmentation was discussed. We learned that there are different categories of segmentation. Each has its own target application. This chapter focused on the network design, implementation, and validation of semantic segmentation.
Our semantic segmentation network was inspired by FCN, which has been the basis of many modern-day, state-of-the-art segmentation algorithms, such as Mask-R-CNN [5]. Our network was further enhanced by ideas from PSPNet, which won first place in the ImageNet 2016 parsing challenge.
Using the VIA labeling tool, a new dataset label for semantic segmentation was generated using the same set of images employed in Chapter 11, Object Detection. The segmentation mask labels all pixels belonging to the same object class.
Our semantic segmentation network was trained and validated using mean IoU and average pixel accuracy metrics. The performance on the test dataset shows that it can effectively classify...
