ARIC: An Activity Recognition Dataset in Classroom Surveillance Images

The Overview of ARIC

ARIC (Activity Recognition In Classroom) is a brand-new and challenging dataset,capturing real classroom scenarios from three different perspectives: front, middle, and rear,

mid perspective	rear perspective	front perspective

each equipped with high-definition 4K cameras. From these videos, images were extracted and annotated to depict the behaviors of students and teachers as the image modality.

Additionally, the audio modality comprises 10 seconds of audio extracted from the videos, 5 seconds before and after each image, we have also preserved the original captions of the images as the text modality.

The dataset, named the ARIC, encompasses classroom-themed scenarios, rich modalities, various perspectives,and an imbalanced class distribution.

Its complexity, diverse behaviors, and crowded scenes present significant challenges.Each modality in the current version consists of 36,453 samples, with a total of 32 annotated behaviors,

covering common behaviors of both teachers and students. The distribution of behaviors can be seen in the below. There is a significant variation in the number of instances for different behaviors.

distribution of activities

Continual Activity Recongnition

For the continual learning evaluation, we present a set of settings of incremental steps,

i.e., the 32 classes are divided into 7 incremental steps and each step contains {8, 4, 4, 4, 4, 4, 4} activity classes,

and you may find more flexible setting in the Readme.md.

Download ARIC Dataset

Usage License: All files in the ARIC dataset can be used for academic purposes only, but any commercial use is prohibited.

To protect the privacy of individuals appearing in the images, we refrain from publishing the original pictures.

Instead, we share annotated class-instance images extracted from the original pictures using pre-trained models' shallow features.

for three pre-trained model, you can choose one and download specified image_features, and you may find more detailed introduction in the Readme.md.

Download: ARIC

The directory structure of the ARIC dataset is as follows:

ARIC-dataset/
            index/
            ├── train/
            │   ├── train_0.txt
            │   ├── train_1.txt
            │   └── ...
            ├── test/
            │   ├── test_0.txt
            │   ├── test_1.txt
            │   └── ...
            image_features/
            ├── vit
            ├── clipvit
            └── resnet
            audio/
            ├── wav_00001.wav
            ├── wav_00002.wav
            └── ...
            text/
            ├── 00001.txt
            ├── 00002.txt
            └── ...
            Readme.md
        

Note for data files

For your convenience in utilizing our dataset, we also provide a dataloader.

It comes with preset task configurations, which can be controlled by passing parameters to specify the data sources and task settings you wish to access.

Due to space limitations, please refer to the ReadMe.md of ARIC-dataset for more details about dataloader.

If you have any questions, please contact us.

Citation

If you find the ARIC dataset useful in your research, please consider citing:

@misc{xu2024aricactivityrecognitiondataset,
      title={ARIC: An Activity Recognition Dataset in Classroom Surveillance Images}, 
      author={Linfeng Xu and Fanman Meng and Qingbo Wu and Lili Pan and Heqian Qiu and Lanxiao Wang and Kailong Chen and Kanglei Geng and Yilei Qian and Haojie Wang and Shuchang Zhou and Shimou Ling and Zejia Liu and Nanlin Chen and Yingjie Xu and Shaoxu Cheng and Bowen Tan and Ziyong Xu and Hongliang Li},
      year={2024},
      eprint={2410.12337},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.12337}, 
}