fjchange / awesome-video-anomaly-detection Goto Github PK

Papers for Video Anomaly Detection, released codes collection, Performance Comparision.

anomalies papers awesome video-anomaly-detection novelty-detection surveillance-videos abnormal-events detection deep-learning

awesome-video-anomaly-detection's Introduction

All good things are on the way ahead.

Hi, I am 冯嘉昌(Jia-Chang Feng).

I join DJI Innovation as a perception algorithm engineer. Besides, I am a Master of SYSU (Sun-Yat-Sen University), supervised by Wei-Shi Zheng.

🔭 Reseach-wise, I mainly focus on:

Video Understanding and Robotics Perception.
More specifically, Video Event Detection and Retrieval, Stereo Vision, Semantic Segmentation, Light-Weight Vision Network.

😄 I am open to:

A job offer with computer vision research and engineering
A PhD position

🌱 Publications:

Cross-modal Consensus Network for Weakly Supervised Temporal Action Localization. ACM MM 2021 paper
MIST: Multiple Instance Self-Training Framework for Video Anomaly Detection. CVPR 2021 paper
Weakly-supervised Action Localization via Hierarchical Mining. Arxiv
Weakly-Supervised Temporal Action Localization by Progressive Complementary Learning Arxiv

📫 Contact me by:

Email (fjchange [AT] hotmail.com)
知乎:「Kiwi」

💬 News:

2022-12-27: I serve as a reviewer for Pattern Recognition (PR).
2022-11-17: I serve as a reviewer for IEEE Transactions on Multimedia (TMM).
2022-11-09: I serve as a reviewer for IEEE Transactions on Circuits and Systems for Video Technology (TSCVT).
2022-07-05: I join DJI as a UAV perception algorithm engineer.
2022-06-24: The internship in ARC Lab of Tencent comes to its end.
2022-06-13: I'm honored with Outstanding Graduate of SYSU.
2022-05-27: A paper about Self-Supervised Visual Pretraining is submitted to NeurIPS 2022.
2022-05-04: I serve as a reviewer for ECCV 2022.
2021-12-31: I serve as a reviewer for PRCV 2022.
2021-12-08: I serve as a reviewer for Neurocomputing.
2021-07-04: A paper about weakly supervised action localization is accepted on ACM MM 2021.
2021-06-29: I serve as a reviewer of PRCV 2021.
2021-06-02: I will start my internship in Applied Research Center Lab (ARC Lab), Platform and Content Group (PCG) of Tencent, to do some research on video understanding and analysis, under the supervision of Ying Shan and Zhongang Qi
2021-05-08: I give a speak on CSIG-Guangdong CVPR Sharing ppt.
2021-04-18: One paper about Video Action Localization is submitted on ACM MM 2021.
2021-04-01: One paper about Video Anomaly Detection is accepted on CVPR 2021. MIST

awesome-video-anomaly-detection's People

Contributors

Stargazers

Watchers

Forkers

ttt749 yangzijia g-morgen secure-and-intelligent-systems-lab mo-white gitcheol frankfan007 yuguangnudt demonzyj56 robinliulovescola xinguozju loicvz190 roc-ng gxlearner dongdem waoup tiebao brngl johnny-lin xuldor ositolam littleserendipity marcinel gityyh robertdigital ckjeter yangyangtiaoguo rongrongtian hzitoun zugexiaodui wanboyang huangdengrong alpha-yang ali-hamid wuqiangch leitang0395 gengcauwong wildfiretm ticks0628 aniketrajpoot darksidesfear rogersguo leiwangai forechoandlook su797 cuitao111 masoudpz csw816 wangjun9276 zmh0124 vivianliangb shuvozitghose yujiangpu20 heleiqiu ysw961015 huyi1998 carlxu121 dinesh98891 augustus-su hoesungryu josepht1962 karthik111 wuzerun-888 mohammad-brdrn easyleeya apanariello4 kwrprojects gekangjian prasaddev9 jimmycreative shanyu-sys josekurian ahmedazab1235 aasimwadood kksf5 qizema hu2375855724 falcon-98 lizhuo97 aspnetcs andysucao manaroqby maingoc1605 mjx8698 jeopardize zcy179 jerry-muyangren ai-reaserchteam aki115 shashwatrajvanshi devtestmrx asadmujtaba

awesome-video-anomaly-detection's Issues

Link is broken(page not find)

HI, @fjchange. Thx for your great work. I have noticed you published new paper which from 2021. But the links of them seem broken.
Thanks for your collection work again!

10-crop and its lable : 关于10-crop提取特征的方式，它的label怎么获得

请问常用的10-crop-I3D特征提取方式，他们的10-crop的lable怎么获得呀。
比如一个异常视频320帧，分为32个instance，每个instance的lable怎么获得。（如果是正常视频，那么32个instance的lable都是0）

The commonly used 10-crop-i3d feature extraction methods and how to obtain their 10 Crop label.
For example, 320 frames of an abnormal video are divided into 32 instances. How can the label of each instance be obtained. (if it is a normal video, the label of 32 instances is 0)

What is the definition of "end-to-end" in the table?

First, thanks for your work, it really help me to better understand the development of VAD in recent year. I encount a question so I want to consult with you.
I note that you mark the "GCN-anomaly"(Zhong 2019) method as "end-to-end". However, in the github provided, I found that extracted features were utilized. And the training process contains 2 stages: generation of persudo labels and training of GCN structure. I wonder why the method is belong to "end-to-end" but the others( like the newest RTFM, which is the last line in table) does not.

MIL VAD question

Hi. Thanks to you, i could research the anomaly detection easily!

I have some question while studying.

I know the MIL refer the "Sultani_Real-World_Anomaly_Detection_CVPR_2018_paper" paper code but what is the VAD??

if above question is correct,
2. how did you get the MIL VAD results by using shanghaiTech data set? because this data set does not have the enough training data set,, (only training data has normal video)
(MIL VAD CVPR 18 Weakly (Re-Organized Dataset) I3D-RGB X 86.3 0.15)

Have a wonderful day.
Thank you

There is a TMAPI 2020 survey

Ramachandra B, Jones M, Vatsavai R R. A survey of single-scene video anomaly detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.

New and better results

https://github.com/wjtan99/BERT_Anomaly_Video_Classification
Can you update your page? Thanks.

It's better to distinguish "Micro-AUC" and "Macro-AUC"

In Table "Performance Comparison on ShanghaiTech", the AUC of SSMT is 90.2 as reported in its paper. However, the authors use the so called "Macro-AUC" (https://github.com/lilygeorgescu/AED-SSMTL#evaluation), which is not the way to calculate AUC in other papers. It's better to distinguish the two metrics.
Also, it is not the first time the authers of SSMT use Macro-AUC. In their previous papers, e.g. "Object-centric Auto-encoders and Dummy Anomalies for Abnormal Event Detection in Video", they use this metric and actually cause unfair comparisons.

How could I get Subway Entrance / Exit dataset?

Hello,Thank you for your excellent collection job.
I'm confused that where can I download Subway Entrance / Exit?
Link (http://vision.eecs.yorku.ca/research/anomalous-behaviour-data/) provides a zip file containing:

GT/
variables.cmd
where-to-get-the-video.txt

Files where-to-get-the-video.txt says:

In order to obtain the video, please contact Amit Adam ([email protected])

Folder GT/ only contains ground truth.

I'm sending a email to Prof. Adam. But I wonder if another way to download this dataset.

Result Value Error

Hi Thank you for your last question answer!
I found the some error.

Sultani.etl | ICME 2020 | Weakly (Re-Organized Dataset) | I3D-RGB | X | 86.3 | 0.15

At the feature column, The AR-Net author who tested the Sultani.etl Shanghai dataset used the only C3D features to train& test the AUC!

This is the part of the paper :
WEAKLY SUPERVISED VIDEO ANOMALY DETECTION VIA CENTER-GUIDED DISCRIMINATIVE LEARNING
"
In order to present a comparison against MIL-based works on ShanghaiTech, we reproduced the method in [9] and adopted the open source code provided by Sultani et al. [7] to conduct the anomaly detection. The above two models are obtained by pretrained C3D.
"

And really thank you for your persistent updates of README.md!
Have a wonderful day!

How to choose the final result to report?

Hi @fjchange, I do benefited a lot from this repo, thanks for your excellent work !
Here i have a confusion when i'm reading the Weakly-Supervised methods based on UCF-CRIME & XD-violence, Both two datasets are only divided into training and testing set, So How did former works choose their final results to report in their paper ?

thank you again!

colab

Any colab versions?