[Paper Review] SCB-Dataset: A Dataset for Detecting Student and Teacher Classroom Behavior

Fan Yang|arXiv (Cornell University)|Apr 5, 2023

Innovative Teaching Methods15 citations

TL;DR

This paper presents SCB-dataset, a public dataset focusing on hand-raising in classroom scenes (11,248 labels across 4,003 images) and evaluates it using YOLOv7 variants, achieving up to 85.3% mAP at 0.5 IoU.

ABSTRACT

Using deep learning methods to detect the classroom behaviors of both students and teachers is an effective way to automatically analyze classroom performance and enhance teaching effectiveness. Then, there is still a scarcity of publicly available high-quality datasets on student-teacher behaviors. We constructed SCB-Dataset a comprehensive dataset of student and teacher classroom behaviors covering 19 classes. SCB-Dataset is divided into two types: Object Detection and Image Classification. The Object Detection part includes 13,330 images and 122,977 labels, and the Image Classification part includes 21,019 images. We conducted benchmark tests on SCB-Dataset using YOLO series algorithms and Large vision-language model. We believe that SCB-Dataset can provide a solid foundation for future applications of artificial intelligence in education. Code:https://github.com/Whiffe/SCB-dataset

Motivation & Objective

• Address the lack of publicly available datasets for student classroom behavior detection.
• Provide a realistic dataset focused on hand-raising in diverse classroom scenarios.
• Demonstrate the viability of object-detection-based approaches (YOLOv7) for classroom behavior analysis.

Proposed method

• Collected real-world classroom images from public sources to capture diverse scenes and challenges.
• Manually annotated hand-raising instances with bounding boxes to create SCB-dataset.
• Trained and evaluated multiple YOLOv7 variants (tiny, standard, X, W6, E6) on SCB-dataset.
• Configured training with 150 epochs, batch size 8, image size 640x640, using a pre-trained backbone.
• Reported precision (P), recall (R), and mean average precision (mAP) at IoU thresholds 0.5 and 0.5:0.95.

Experimental results

Research questions

• RQ1 How well can YOLOv7-based detectors identify hand-raising behavior in real classroom images?
• RQ2 What is the impact of model size and architecture within YOLOv7 variants on detection performance for classroom behaviors?
• RQ3 What are the dataset characteristics and challenges (scale variation, occlusion, viewpoint) affecting detection in classrooms?

Key findings

• SCB-dataset contains 11,248 labels across 4,003 images focusing on hand-raising in classrooms.
• YOLOv7-tiny achieves P=75.7%, R=75.9%, mAP@0.5=79.7%, mAP@0.5:0.95=54%.
• YOLOv7 (standard) achieves P=83.1%, R=80.6%, mAP@0.5=84.9%, mAP@0.5:0.95=61.8%.
• YOLOv7-X achieves P=81.4%, R=81.4%, mAP@0.5=84.6%, mAP@0.5:0.95=62.3%.
• YOLOv7-W6 achieves P=84.1%, R=79.0%, mAP@0.5=85.3%, mAP@0.5:0.95=62.1%.
• YOLOv7-E6 achieves P=82.2%, R=80.5%, mAP@0.5=85.1%, mAP@0.5:0.95=63.1%.