Like human drivers, self-driving cars need to understand their environment and be aware of the elements around them. Applying semantic segmentation to the video images from a front camera would allow the system to know whether other cars are around, to know whether pedestrians or bikes are crossing the road, to follow traffic lines and signs, and more.
This is, therefore, a critical process, and researchers are putting in lots of effort into refining the models. For that reason, multiple related datasets and benchmarks are available. The Cityscapes dataset (https://www.cityscapes-dataset.com) we chose for our demonstration is one of the most famous. Shared by Marius Cordts et al. (refer to The Cityscapes Dataset for Semantic Urban Scene Understanding, Proceedings of the IEEE CVPR Conference), it contains video sequences from multiple cities, with semantic labels for more than 19 classes (road, car, plant, and so on). A notebook is specifically dedicated to...
