VidProUCF

VidProUCF is a video processing system to extract and store features from videos of the UCF101 dataset. It comes with an interactive web application to filter videos based on the extracted features.

Demo

A demo app built on a tiny subset of the UCF101 dataset is hosted at https://vidpro-ucf-iamarijit.streamlit.app/.

Setup

1. Install docker using instructions at https://docs.docker.com/engine/install/
2. Build docker image

./env/build.sh

3. Start docker container

./env/start.sh

4. Open container CLI

./env/bash.sh

5. Install dependencies

pip install -r requirements.txt

Usage

Download dataset

wget  --no-check-certificate  https://www.crcv.ucf.edu/datasets/human-actions/ucf101/UCF101.rar
unrar  x  UCF101.rar

Convert the raw videos to H264 codec which is supported by HTML5 video player

./convert.sh

Note: Change video_dir to UCF101 dataset path on local machine

Extract & store features

python process.py

Note: Change root to UCF101 dataset path on local machine

Run web app to filter videos

streamlit run deploy.py

Open http://localhost:8501 on your browser