In semantic segmentation, what is the pixel-level classification method used to differentiate between drivable road area and non-drivable obstacles?

In semantic segmentation for autonomous driving, the primary method used to differentiate between drivable road area and obstacles is the Fully Convolutional Network, or FCN. This architecture processes an entire image to produce a pixel-wise label map, meaning every single pixel in the input image is assigned a category, such as road, vehicle, pedestrian, or background. Unlike traditional classification that outputs a single label for an entire image, the FCN replaces the final fully connected laye....