Introduction

This repository is for extracting C3D or I3D features from videos.

Requirements

• Python 3
• CUDA
• numpy
• tqdm
• PyTorch (1.2)
• torchvision
  Recommend: the environment can be established by run

conda env create -f anomaly_feature.yaml

Video preprocessing

1. Using Videos_to_frames_multiprocessing.py to generate frames from videos.

ps: if you want to extract I3D feature, it is recommend that using denseflow (https://github.com/open-mmlab/denseflow) to generate RGB and Optical_flow images.

Input list generation

1. Using the ./dataset/write_data_label_txt_new.py

generate the index file for running feature_extract.py.

Download the model.zip in (link: https://pan.baidu.com/s/1g4XGLqRoRJhQwIGtHif2jg password: dzqm), and unpress it to the root.

Feature extraction

1. Using feature_extract.py to generate I3D or C3D features.

Using script

python feature_extract.py --dataset shanghaitech --modelName i3d --snapshot ./model/i3d/i3d_model_weight/model_kinetics_rgb.pth --datamodal rgb

to extract i3d rgb features.

Using script

python feature_extract.py --dataset shanghaitech --modelName i3d --snapshot ./model/i3d/i3d_model_weight/model_kinetics_flow.pth --datamodal flow

to extract i3d flow features.

Using script

python feature_extract.py --dataset shanghaitech --modelName c3d --snapshot ./model/c3d/c3d.pickle --datamodal rgb

to extract c3d features.

Using dataset_creater.py to generate final feature file.