/Anomaly-Detection

Anomaly detection (Pytorch)

Paper: https://arxiv.org/pdf/1801.04264v3

p9

Implemetation:

1- Clone this github repo:

git clone https://github.com/ekosman/AnomalyDetectionCVPR2018-Pytorch.git

2- Setup the environment to run this code:

cd to AnomalyDetectionCVPR2018-Pytorch

You will see environment.yaml file. Open it. You need to install all these modules. I am not working on anaconda, so I can’t use environment.yaml as it is. You can convert this environment.yaml into requirements.txt or just all these modules 1 by 1 through command line using pip. Eg:

pip install torch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 --index-url https://download.pytorch.org/whl/cu118

pip install numpy

pip install opencv-python

pip install matplotlib

pip install pandas

pip install av

Note: you have to check the cuda version installed in your machine and then pick the right command to install pytorch. IN my case I have installed the pytorch with cuda 11.8

You need to install LAV filters also.

Visit this page https://github.com/Nevcairiel/LAVFilters/releases and download LAVFilters-0.79.2-Installer.exe which is under assets and then install it.

DATASET:

The dataset can be also downloaded from the following link: https://visionlab.uncc.edu/download/summary/60-data/477-ucf-anomaly-detection-dataset

You can also download dataset in parts through following link

https://www.dropbox.com/sh/75v5ehq4cdg5g5g/AABvnJSwZI7zXb8_myBA0CLHa?dl=0

data

Feature Extraction model:

Download the pretrained C3D weights (Sports1M) from here: http://imagelab.ing.unimore.it/files/c3d_pytorch/c3d.pickle

Paste it in pretrained folder and then run this command:

python video_demo.py --feature_extractor "pretrained/c3d.pickle" --feature_method "c3d" --ad_model "D:\anomaly_detection\AnomalyDetectionCVPR2018-Pytorch\exps\c3d\models\epoch_80000.pt"