Face recognition has evolved as a prominent biometric authentication\nmodality. However, vulnerability to presentation attacks curtails its reliable\ndeployment. Automatic detection of presentation attacks is essential for secure\nuse of face recognition technology in unattended scenarios. In this work, we\nintroduce a Convolutional Neural Network (CNN) based framework for presentation\nattack detection, with deep pixel-wise supervision. The framework uses only\nframe level information making it suitable for deployment in smart devices with\nminimal computational and time overhead. We demonstrate the effectiveness of\nthe proposed approach in public datasets for both intra as well as\ncross-dataset experiments. The proposed approach achieves an HTER of 0% in\nReplay Mobile dataset and an ACER of 0.42% in Protocol-1 of OULU dataset\noutperforming state of the art methods.\n