Video anomaly detection data set (shanghaitech)

Inge2022-06-23 18:04:12

List of articles

1 introduce

source address https://svip-lab.github.io/dataset/campus_dataset.html
Address of thesis https://openaccess.thecvf.com/content_cvpr_2018/papers/Liu_Future_Frame_Prediction_CVPR_2018_paper.pdf
Data sets and code https://github.com/StevenLiuWen/ano_pred_cvpr2018

2 summary

A direct goal of anomaly detection model training is that it can be directly applied to multi view and multi scene . However , Almost all Existing datasets only contain videos taken with a fixed angle camera , And lack of diversity of scenes and perspectives . To increase scene diversity , A new anomaly detection data set is constructed ShanghaiTech. Besides , An anomaly caused by sudden motion is introduced in this data set , Such as chasing and fighting . These exceptions are not available in existing datasets , Make the data set more suitable for the real scene .
The anomaly detection data set is summarized as follows :
1)CUHK Avenue: contain 16 Training videos and 21 Test videos , common 47 Abnormal events , Contains throwing objects 、 Stroll and run . The size of a person changes with the position and angle of the camera ;
2)Pedestrian 1 (Ped1): contain 34 Training videos and 36 Test videos , It includes 40 An irregular event . All these exceptions are related to bicycles 、 Cars and other means of transportation .
3)Pedestrian 2 (Ped2): contain 16 Training videos and 12 Test videos , contain 12 Abnormal events .Ped2 The definition of exception and Ped1 identical .
4)Subway: There are two types of , I.e. inlet and outlet . Unusual events include going in the wrong direction and wandering . This data set is recorded in an indoor environment , The above data are recorded in the outdoor environment .
5)ShanghaiTech: contain 13 A scenario , It has complex lighting conditions and camera angle . contain 130 Exceptional events and more than 270000 Training frames . Besides , Pixel level annotation of abnormal events is given .

3 Bib

@inproceedings{
Liu:2018:65366545,
author = {
Wen Liu and Wei Xin Luo and Dong Ze Lian and Sheng Hua Gao},
title = {
Future frame prediction for anomaly detection--a new baseline},
booktitle = {
{
IEEE} conference on Computer Vision and Pattern Recognition},
pages = {
6536--6545},
year = {
2018}
}

thank
Similar articles