1 SCB-dataset

Student Classroom Behavior dataset

Paper Name	Dataset and Trained weights Link	class
Student Classroom Behavior Detection based on Improved YOLOv7	Baidu Netdisk dataset extraction code: kjek Baidu Netdisk models extraction code: epvk	hand-raising, reading, writing
SCB-dataset	Baidu Netdisk Dataset & Models extraction code：SCB5 huggingface

Baidu Netdisk Dataset & Models extraction code：qg8k

This is the data of SCB-Dataset3, which includes university classroom data. However, we are remaking it because the previous data had issues such as insufficient quantity and single scenarios.

2 Contact

If you encounter any issues with the SCB Dataset or cooperation, please feel free to contact me via email at : [email protected]

3 YOLO + CrowdHuman

4 Commercial Use Restrictions

Please note that this project (including but not limited to code, datasets, model weights, etc.) is intended for academic research, personal learning, and non-commercial use only. Any commercial use (including but not limited to commercial software development, commercial data services, commercial product development, etc.) is strictly prohibited without the explicit written permission of the copyright holder. Any unauthorized commercial use will constitute an infringement of the copyright holder's rights, and the copyright holder reserves the right to pursue legal action. If you have a need for commercial use, please contact the copyright holder for authorization through the following means:

• Contact Email: [email protected] The copyright holder will decide whether to grant commercial use permissions based on the specific circumstances and may require the signing of a corresponding authorization agreement.

5 Copyright Statement

All content of this project, including but not limited to code, datasets, model weights, documents, etc., is protected by copyright. The copyright holder reserves all rights. Without authorization, no individual or organization is permitted to copy, distribute, modify, or otherwise use the content of this project unless explicitly granted permission by the copyright holder. Thank you for respecting and supporting the rights of the copyright holder.

Citation

Please cite the following paper if you use our dataset.

@article{yang2023scb,
  title={SCB-dataset: A dataset for detecting student classroom behavior},
  author={Yang, Fan},
  journal={arXiv preprint arXiv:2304.02488},
  year={2023}
}

6 Papers and Software copyright

6.1 Papers

29 October 2023, ICIG 2023 , Student Classroom Behavior Detection Based on YOLOv7+BRA and Multi-model Fusion

杨帆,詹泽慧.基于RT-DETR-ASF的学生科学探究实验行为检测研究[J].数字教育,2024,10(05):14-23.

6.2 Software copyright

7 Acknowledgements

For the development of this dataset, we would like to thank the following individuals, companies and universities for their support:

ELEMENT TECH（成都元素科技有限公司 https://www.elementech.net/）, SICHUAN UNIVERSITY OF SCIENCE AND TECHNOLOGY, Wang Wengang（四川科技职业技术学院 王文刚）, Urban Vocational College Of Sichuan, Li Jianlong（四川城市职业学院 李健龙）, Chengdu Neusoft University（成都东软学院）, Sichuan Normal University（四川师范大学）, Beijing Normal University（北京师范大学）

8 Open Source Dataset

8.1 STBD-08

Thanks to the CBPH-Net https://github.com/icedle/CBPH-Net authors for their contributions, here is the download link for the CBPH-Net contribution dataset STBD-08:

BiTNet: A lightweight object detection network for real-time classroom behavior recognition with transformer and bi-directional pyramid network

CBPH-Net: A Small Object Detector for Behavior Recognition in Classroom Scenarios

The STBD-08 paper claims that the dataset includes 4432 images and 151574 annotations. Nevertheless, when we calculated the data provided by the author, we discovered that the actual quantities are much larger than these figures. Our statistics indicate that the dataset has 8884 images (with 7052 in the training set and 1,832 in the validation set) and 267,888 annotations (including 212,728 in the training set and 55,160 in the validation set).

Through online search, we found that the STBD-08 dataset is completed based on the dataset publicly sale online (the dataset also has 8,884 pieces), and the data volume is far lower than that of the dataset publicly sold online.

However, when we cleaned the STBD-08 dataset, we found that there are still many problematic data in it, such as non-standard bounding boxes (bbox) and some class labeling errors.

In other words, the author of STBD-08 has not made the dataset they created public, and only the original dataset purchased online is disclosed.

8.2 A dataset of student classroom behavior from a paid website

This dataset is highly similar to STBD-08, mainly in that the images are highly similar and the quality of annotations is also poor

8.3 universe roboflow website

8.4 ClaBehavior

The ClaBehavior paper mentions 1,342 images and 9,911 annotations. However, we actually found only 400 images and 8,083 annotations on GitHub, including:

• Train Dataset: 360 images and 7,250 annotations
• Val Dataset: 40 images and 833 annotations

The categories include: Write, Read, Lookup, Turn_head, Raise_hand, Stand, Discuss.

8.5 SCBehavior

The SCBehavior paper mentions that there are 1346 images. However, when we checked the author's GitHub, we found only 400 damaged images that cannot be viewed (360 in the Train Dataset and 40 in the Val Dataset).

8.6 UK_Datasets

UK_Datasets paper is derived from the 2019 elementary school classroom videos collected from the National Education Resources Public Service Platform (NERPSP).

UK_Datasets extracted 8754 images by frame, and considering the detection needs in real classroom scenarios, it classified these images into eight categories of typical student behaviors: writing, reading, listening, raising hands, turning, standing, discussing, and accepting teacher instructions.

The author categorized the test set portions of UK_Datasets according to the degree of occlusion: "Heavy Occlusion (HO)" and "Low Occlusion (LO)".

Unfortunately, when we downloaded the UK_Datasets for statistics, we found that the data was not original. Specifically, it originated from the 2.1 STBD-08 section and the dataset publicly available online as introduced in this paper. The author merely divided and counted these existing datasets.

Since the data itself is plagiarized and not original, this paper will not conduct statistical analysis on its data.

8.7 BNU-Wu Student-Class-Behavior-Dataset

not open source

9 Close Source Dataset

Star History