In this research paper, we propose an attendance system that uses deep learning models to automate the attendance marking process in educational institutions. A professor would be able to log in to the system, upload or record the classroom video using their mobile device. It will use face detection and recognition techniques to accurately count the students and mark their attendance. Our focus is on face detection, using the Retina-Face model due to its high accuracy and ability to detect faces in varied conditions such as occlusions, different angles, lighting scenarios, and face alignment. The performance of the model is tested using both pre-existing images and classroom data along with real-time videos. Results demonstrate promising accuracy in detecting faces and counting heads, which ensures the scalability of the system for deployment and usage throughout the university. Accuracy of 99.375% and 99.19% was achieved in real-time image and video, respectively. This paper discusses the architecture and implementation of the system along with face detection techniques, focusing on accuracy as a foundation for automated attendance systems.