SEMANTIC SEGMENTATION FOR SELF DRIVING CARS

This paper proposes a novel approach for Semantic segmentation which is one of the biggest challenge increasing in an order and have been making humans hold keen active interest to result in fast and accurate semantic segmentation. Whereas At present, we are trying to solve this problem of semantic segmentation using the segnet which makes its more accurate interms of accuracy, computational time, and inference time. and here we are using segnet model to take this to the next level which includes max-pooling, Batch normalization techniques to map low-resolution features to input resolution for pixelwise classification and the architecture here consists of an encoder which takes the input image and is identical to 13 convolutional layers and a decoder that uses segnet followed by pixel-wise classification layer. and also when compared with other architectures segnet provides good performance with competitive inference time and most efficient memory. So, therefore here we are presenting deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet.