Crowd counting problem is a challenging task in computer vision and image analysis. It has many applications in the real world such as crowd management, public safety, and urban planning. Our proposal in this paper is a mask-aware transformer-based network for crowd counting that uses the background/foreground mask information to improve density regression accuracy. Our backbone network is a pyramid vision transformer. Our proposed Mask-aware transformer (M-Trans) takes into consideration the background/foreground mask information. We further improve the performance by applying a greedy ensemble strategy. Our experimental evaluation shows that our mask-aware network achieves state-of-the-art performance on standard benchmarking datasets for crowd counting such as ShanghaiTech and UCF-QNRF datasets.