A Dynamic Approach for Railway Semantic Segmentation

"Railway semantic segmentation is the task of highlighting rail blades in images taken from ego-view train. Solving this allows for further image processing on rails, which can be used more complex problems such as switch or fault detection. In paper we approach railway using two deep architectures U-Net family, and ResUNet++, most comprehensive dataset available at time writing scene, namely RailSem19. We also investigate effects augmentations different training sizes, well performance models dark images. have compared our solution to other approaches obtained competitive results with larger scores. Keywords phrases: Binary Semantic Segmentation, Encoder-decoder, Railway blades, Deep Learning. "