AIRC-SMARTCLASS: A Multi-Modal Classroom Dataset for Automatically Controlling Devices in Smart Classroom Part 1

Collected at the Computer Laboratory, Information Technology - Library Center of the University of Transport Technology in Hanoi, Vietnam, the dataset captures real-world classroom interactions among 249 participants (5 lecturers and 244 students). The visual recordings were acquired using a single wall-mounted, wide-angle surveillance camera installed at ceiling height to provide a top-down longitudinal perspective.

In total, the collection contains 45 original video clips in .avi formats, from which 12,429 Full-HD (1920 x 1080) image frames were sampled at a rate of 1 frame per 3 seconds. To support multi-task deep learning research, the dataset provides standardized annotations in YOLO Ultralytics Detection(.txt) formats, generated through semi-automated labeling in CVAT and manually verified by experts. These annotations span three core computer vision tasks: people counting with 230,409 full-body bounding boxes, face detection and identification with 62,001 face bounding boxes, and air conditioner screen with 12,429 localized display bounding boxes structured to extract operational state fields including STATUS, TEMP, MODE, and FAN.