Lane detection using FCN by Semantic Segmentation

FCN : FCN is one of the first proposed models for end-to-end semantic segmentation. Here standard image classification models such as VGG and AlexNet are converted to fully convolutional by making FC layers 1x1 convolutions. At FCN, transposed convolutions are used to upsample, unlike other approaches where mathematical interpolations are used. The three variants are FCN8, FCN16 and FCN32. In FCN8 and FCN16, skip connections are used.