Signal reconstruction is a challenging aspect of computational imaging as it\noften involves solving ill-posed inverse problems. Recently, deep feed-forward\nneural networks have led to state-of-the-art results in solving various inverse\nimaging problems. However, being task specific, these networks have to be\nlearned for each inverse problem. On the other hand, a more flexible approach\nwould be to learn a deep generative model once and then use it as a signal\nprior for solving various inverse problems. We show that among the various\nstate of the art deep generative models, autoregressive models are especially\nsuitable for our purpose for the following reasons. First, they explicitly\nmodel the pixel level dependencies and hence are capable of reconstructing\nlow-level details such as texture patterns and edges better. Second, they\nprovide an explicit expression for the image prior which can then be used for\nMAP based inference along with the forward model. Third, they can model long\nrange dependencies in images which make them ideal for handling global\nmultiplexing as encountered in various compressive imaging systems. We\ndemonstrate the efficacy of our proposed approach in solving three\ncomputational imaging problems: Single Pixel Camera (SPC), LiSens and FlatCam.\nFor both real and simulated cases, we obtain better reconstructions than the\nstate-of-the-art methods in terms of perceptual and quantitative metrics.\n