Despite the growing success of Convolution neural networks (CNN) in the\nrecent past in the task of scene segmentation, the standard models lack some of\nthe important features that might result in sub-optimal segmentation outputs.\nThe widely used encoder-decoder architecture extracts and uses several\nredundant and low-level features at different steps and different scales. Also,\nthese networks fail to map the long-range dependencies of local features, which\nresults in discriminative feature maps corresponding to each semantic class in\nthe resulting segmented image. In this paper, we propose a novel multi-scale\nattention network for scene segmentation purposes by using the rich contextual\ninformation from an image. Different from the original UNet architecture we\nhave used attention gates which take the features from the encoder and the\noutput of the pyramid pool as input and produced out-put is further\nconcatenated with the up-sampled output of the previous pyramid-pool layer and\nmapped to the next subsequent layer. This network can map local features with\ntheir global counterparts with improved accuracy and emphasize on\ndiscriminative image regions by focusing on relevant local features only. We\nalso propose a compound loss function by optimizing the IoU loss and fusing\nDice Loss and Weighted Cross-entropy loss with it to achieve an optimal\nsolution at a faster convergence rate. We have evaluated our model on two\nstandard datasets named PascalVOC2012 and ADE20k and was able to achieve mean\nIoU of 79.88% and 44.88% on the two datasets respectively, and compared our\nresult with the widely known models to prove the superiority of our model over\nthem.\n