NWPU Campus dataset

Congqi Cao   Yue Lu   Peng Wang   Yanning Zhang  

National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology (ASGO), School of Computer Science, Northwestern Polytechnical University, China

Paper: A New Comprehensive Benchmark for Semi-supervised Video Anomaly Detection and Anticipation

NWPU Campus is a dataset proposed for (semi-supervised) video anomaly detection (VAD) and video anomaly anticipation (VAA). It is currently the largest and most complex dataset in its field with 43 scenes, 28 classes of anomalous events and 16 hours of videos. Especially, it contains scene-dependent anomalies, which means an event may be normal in one scene but abnormal in another.

Introduction

A number of normal and anomalous events in different scenes are displayed in Figure 1. All the 28 classes of anomalous events are listed in Table 1.
Table 2 and Figure 2 show the statistics of the dataset, including durations and distributions.

Samples

fig1.jpg
Figure 1. Samples from the proposed NWPU Campus dataset. The samples in the first column are normal events, while the others are different types of anomalous events.

Classes of anomalies

Table 1. The list of anomaly classes in NWPU Campus dataset. "s.d." stands for a scene-dependent anomaly.
Climbing fence Car crossing square Cycling on footpath (s.d.) Kicking trash can
Jaywalking Snatching bag Crossing lawn Wrong turn (s.d.)
Cycling on square Chasing Loitering Scuffle
Littering Forgetting backpack U-turn Battering
Driving on wrong side Falling Suddenly stopping cycling in the middle of the road Group conflict
Climbing tree Stealing Illegal parking Trucks (s.d.)
Protest Playing with water Photographing in restricted area (s.d.) Dogs

Statistics

Table 2. Frame count and duration of the NWPU Campus dataset.
NWPU Campus (25 FPS)
1,466,073 (16.29h)
Training frames Testing frames
1,082,014 (12.02h) 384,059 (4.27h)
Normal Normal Abnormal
1,082,014 (12.02h) 318,793 (3.54h) 65,266 (0.73h)


fig2.jpg
Figure 2. The distributions of training and testing videos according to duration (a),
and abnormal testing videos according to the percentage of abnormal frames in each video (b).

Download the dataset

Terms & Conditions of use

The NWPU Campus dataset is released for academic research only, and is free to researchers from educational or research institutes for non-commercial purposes. The use of the NWPU Campus dataset is governed by the following terms and conditions:

We hold no liability for any undesirable consequences of using the dataset. All rights of the NWPU Campus dataset are reserved.

Download

There are 305 training videos and 242 testing videos in the NWPU Campus dataset. All of the videos amount to 76.6 GB in disk space.

Please note that most of the abnormal behaviors in the dataset are performed with careful protection. Do not mimic these behaviors in reality.

Download from BaiduYun (code: il54) [101 downloads] or Google Drive

Code

Github

Citation

	@InProceedings{Cao_2023_CVPR,
	    author    = {Cao, Congqi and Lu, Yue and Wang, Peng and Zhang, Yanning},
	    title     = {A New Comprehensive Benchmark for Semi-Supervised Video Anomaly Detection and Anticipation},
	    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
	    month     = {June},
	    year      = {2023},
	    pages     = {20392-20401}
	}
			
Paper link

Contact

For further questions and suggestions, please contact Yue Lu (zugexiaodui@mail.nwpu.edu.cn).

Visits of this page: