In this paper, we present a novel neural network using multi scale feature\nfusion at various scales for accurate and efficient semantic image\nsegmentation. We used ResNet based feature extractor, dilated convolutional\nlayers in downsampling part, atrous convolutional layers in the upsampling part\nand used concat operation to merge them. A new attention module is proposed to\nencode more contextual information and enhance the receptive field of the\nnetwork. We present an in depth theoretical analysis of our network with\ntraining and optimization details. Our network was trained and tested on the\nCamvid dataset and Cityscapes dataset using mean accuracy per class and\nIntersection Over Union (IOU) as the evaluation metrics. Our model outperforms\nprevious state of the art methods on semantic segmentation achieving mean IOU\nvalue of 74.12 while running at >100 FPS.\n