InDiP: Intelligent Digital Painter for Low Resolution Face Inpainting

In the recent years Artificial Intelligence (AI) has become a go to approach to solve any kind of real-world problems. In these pandemic times, there are many issues arising around the world. One of such issues is the identification of a masked person. It has become difficult to even recognize the very well-known persons due to masks, which has made the field of surveillance suffer a lot. The Object of this research is to solve this issue by reconstructing the face behind the mask using deep learning models. We are proposing a network called as Intelligent Design Painter (InDiP) with capabilities of reconstructing the face behind the mask at low resolutions. This would make the recognition of the masked persons easier. To achieve this target initially work has been done on reconstructing the general human images for fine tuning. After fine tuning the algorithm it has applied on masked faces to obtain the desired results. When checked the accuracy with the original image, the results seem satisfactory. In this approach a sequence Transformer is trained to forecast pixels based on data from a succession of 2D inputs without taking into consideration any of the 2D input structure. Even though the GPT-2 scale model has been trained on low-resolution ImageNet datasets without labels, the network is able to perform satisfactorily. So by using this model ONE can deduce the facial images of the masked people.