An information retrieval system is built to provide users with the most relevant information in a timely and efficient manner and it consists of many stages: data collection, data preprocessing, indexing, data retrieving, ranking, and presentation. In this report, we use a machine learning-based information retrieval system designed for video event retrieval. The system uses the combination of CNNs, ViT, and CLIP in feature extraction, and the ScaNN method in ranking proved effective in providing accurate and efficient search results.
Ubuntu 18.04.6
Open CLI, run
cd backend
python3 datacollecting.py
Excute
pip install -r requirements.txt
uvicorn main:app
Open new CLI (still open the remaining CLI), excute
cd frontend
npm install
npm start
Web system displays on port 4000
Retrieve Video by Vietnamese description or English description. Detail can be used to view all extracted keyframes with the video. KNN to find the keyframes with the closest distance to the selected keyframe.