Cross Epoch Learning for Weakly Supervised Anomaly Detection in Surveillance Videos

official implement code for our paper: Cross-Epoch Learning for Weakly Supervised Anomaly Detection in Surveillance Videos

Tabel of Contents

• Installation
• Data Preparation
• Getting Started

Installation

Requirements

• CUDA 10.1
• Python=3.6
• PyTorch=1.4.0
• torchvision=0.4.2
• fvcore
• simplejson
• opencv-python

Data Preparation

Shanghai Tech

ShanghaiTech is a medium-scale anomaly detection dataset, including 437 videos. The re-split for weakly supervised task is from Graph convolutional label noise cleaner

UCF-Crime

UCF-Crime a large-scale complex dataset for anomaly detection. It contains 13 real-world anomalous behaviors, distributed in 1,900 untrimmed videos with a total duration of 128 hours.

Getting Started

step 1

Please make the video data to feature data via C3D or I3D we also release a simple implement for anomaly feature extractor code page

step 2

Before train the model, please check the hyper-parameter file in ./net/config/defaults.py and config/xxx.yaml run the train.py if the model,dataset and hyper-parameter is all already.

stpe 3

Inference the mode via inference.py and eval_auc_xxx.py