The invention discloses a classroom quality bi-directional evaluation system based on depth learning, which comprises a classroom terminal, a teacher terminal, an educational administration terminal,a student terminal and a cloud server. The teacher terminal collects, processes and sends classroom pictures and videos. The teacher terminal sends text information and images to the cloud server to release the job, notify the examination time and place, and receive and display the analysis results. The educational administration terminal receives and displays the basic information of each teacher's class, history teaching situation and students' learning situation; The student terminal receives and displays the notification of class assignment, the notification of examination time and place,the individual classroom performance and the teaching quality evaluation results of each teacher. Based on the received classroom pictures and frames, the cloud server uses the object detection model,face detection model and face segmentation model to analyze the attendance, the listening rate, classroom behaviors, classroom expression, classroom seating and feedback analysis results.